PyFITS Documentation


The PyFITS module is a Python library providing access to FITS files. FITS (Flexible Image Transport System) is a portable file standard widely used in the astronomy community to store images and tables.


PyFITS requires Python version 2.6 or newer. Support for Python 3.0 through 3.2 is deprecated, but Python 3.3 and above are fully supported. PyFITS also requires the numpy package. Information about numpy can be found at:

PyFITS’ source code is mostly pure Python, but includes an optional C module which wraps CFITSIO for compression support. The latest source distributions and binary installers for Windows can be downloaded from:

Or from the Python Package Index (PyPI) at:

PyFITS can be installed using pip (the recommended tool for installing Python packages) using:

$ pip install pyfits

To install PyFITS manually from the source distribution, unpack the tar file and type:

pip install .

This will install PyFITS in the system’s Python site-packages directory. If your system permissions do not allow this kind of installation, use of virtualenv for personal installations is recommended.

In this guide, we’ll assume that the reader has basic familiarity with Python. Familiarity with numpy is not required, but it will help to understand the data structures in PyFITS.

User Support

The official PyFITS web page is:

If you have any question or comment regarding PyFITS, user support is available through the STScI Help Desk:

* E-mail:
* Phone: (410) 338-1082

Quick Tutorial

This section provides a quick introduction of using PyFITS. The goal is to demonstrate PyFITS’ basic features without getting into too much detail. If you are a first time user or an occasional PyFITS user, using only the most basic functionality, this is where you should start. Otherwise, it is safe to skip this chapter.

After installing numpy and PyFITS, start Python and load the PyFITS library. Note that the module name is all lower case.

>>> import pyfits

Reading and Updating Existing FITS Files

Opening a FITS file

Once the PyFITS module is loaded, we can open an existing FITS file:

>>> hdulist ='input.fits')

The open() function has several optional arguments which will be discussed in a later chapter. The default mode, as in the above example, is “readonly”. The open method returns a PyFITS object called an HDUList which is a list-like object, consisting of HDU objects. An HDU (Header Data Unit) is the highest level component of the FITS file structure, consisting of a header and (typically) a data array or table.

After the above open call, hdulist[0] is the primary HDU, hdulist[1] is the first extension HDU (if there are any extensions), and so. It should be noted that PyFITS is using zero-based indexing when referring to HDUs and header cards, though the FITS standard (which was designed with FORTRAN in mind) uses one-based indexing.

The HDUList has a useful method, which summarizes the content of the opened FITS file:

Filename: test1.fits
No. Name  Type       Cards Dimensions Format
0 PRIMARY PrimaryHDU   220 ()         int16
1 SCI     ImageHDU      61 (800, 800) float32
2 SCI     ImageHDU      61 (800, 800) float32
3 SCI     ImageHDU      61 (800, 800) float32
4 SCI     ImageHDU      61 (800, 800) float32

After you are done with the opened file, close it with the HDUList.close() method:

>>> hdulist.close()

The headers will still be accessible after the HDUList is closed. The data may or may not be accessible depending on whether the data are touched and if they are memory-mapped, see later chapters for detail.

Working with large files

The function supports a memmap=True argument that allows the array data of each HDU to be accessed with mmap, rather than being read into memory all at once. This is particularly useful for working with very large arrays that cannot fit entirely into physical memory. memmap=True is the default value as of PyFITS v3.1.0.

This has minimal impact on smaller files as well, though some operations, such as reading the array data sequentially, may incur some additional overhead. On 32-bit systems arrays larger than 2-3 GB cannot be mmap’d (which is fine, because by that point you’re likely to run out of physical memory anyways), but 64-bit systems are much less limited in this respect.


When opening a file with memmap=True, because of how mmap works this means that when the HDU data is accessed (i.e. hdul[0].data) another handle to the FITS file is opened by mmap. This means that even after calling hdul.close() the mmap still holds an open handle to the data so that it can still be accessed by unwary programs that were built with the assumption that the .data attribute has all the data in-memory.

In order to force the mmap to close either wait for the containing HDUList object to go out of scope, or manually call del hdul[0].data (this works so long as there are no other references held to the data array).

Unsigned integers

Due to the FITS format’s FORTRAN origins, FITS does not natively support unsigned integer data in images or tables. However, there is a common convention to store unsigned integers as signed integers, along with a shift instruction (a BZERO keyword with value 2 ** (BITPIX - 1)) to shift up all signed integers to unsigned inters. For example, when writing the value 0 as an unsigned 32-bit integer, it is stored in the FITS file as -32768, along with the header keyword BZERO = 32768.

PyFITS recognizes and applies this convention by default, so that all data that looks like it should be interpreted as unsigned integers is automatically converted (this applies to both images and tables). In PyFITS versions prior to v3.4.0 this was not applied automatically, and it is necessary to pass the argument uint=True to open(). In v3.4.0 or later this is the default.

Even with uint=False, the BZERO shift is still applied, but the returned array is of “float64” type. To disable scaling/shifting entirely, use do_not_scale_image_data=True.

Working with compressed files

The open() function will seamlessly open FITS files that have been compressed with gzip, bzip2 or pkzip. Note that in this context we’re talking about a fits file that has been compressed with one of these utilities - e.g. a .fits.gz file. Files that use compressed HDUs within the FITS file are discussed in Compressed Image Data.

There are some limitations with working with compressed files. For example with Zip files that contain multiple compressed files, only the first file will be accessible. Also bzip does not support the append or update access modes.

When writing a file (e.g. with the writeto() function), compression will be determined based on the filename extension given, or the compression used in a pre-existing file that is being written to.

Working with FITS Headers

As mentioned earlier, each element of an HDUList is an HDU object with .header and .data attributes, which can be used to access the header and data portions of the HDU.

For those unfamiliar with FITS headers, they consist of a list of 80 byte “cards”, where a card contains a keyword, a value, and a comment. The keyword and comment must both be strings, whereas the value can be a string or an integer, floating point number, complex number, or True/False. Keywords are usually unique within a header, except in a few special cases.

The header attribute is a Header instance, another PyFITS object. To get the value associated with a header keyword, simply do (a la Python dicts):

>>> hdulist[0].header['targname']

to get the value of the keyword targname, which is a string ‘NGC121’.

Although keyword names are always in upper case inside the FITS file, specifying a keyword name with PyFITS is case-insensitive, for the user’s convenience. If the specified keyword name does not exist, it will raise a KeyError exception.

We can also get the keyword value by indexing (a la Python lists):

>>> hdulist[0].header[27]

This example returns the 28th (like Python lists, it is 0-indexed) keyword’s value–an integer–96.

Similarly, it is easy to update a keyword’s value in PyFITS, either through keyword name or index:

>>> prihdr = hdulist[0].header
>>> prihdr['targname'] = 'NGC121-a'
>>> prihdr[27] = 99

Please note however that almost all application code should update header values via their keyword name and not via their positional index. This is because most FITS keywords may appear at any position in the header.

It is also possible to update both the value and comment associated with a keyword by assigning them as a tuple:

>>> prihdr = hdulist[0].header
>>> prihdr['targname'] = ('NGC121-a', 'the observation target')
>>> prihdr['targname']
>>> prihdr.comments['targname']
'the observation target'

Like a dict, one may also use the above syntax to add a new keyword/value pair (and optionally a comment as well). In this case the new card is appended to the end of the header (unless it’s a commentary keyword such as COMMENT or HISTORY, in which case it is appended after the last card with that keyword).

Another way to either update an existing card or append a new one is to use the Header.set() method:

>>> prihdr.set('observer', 'Edwin Hubble')

Comment or history records are added like normal cards, though in their case a new card is always created, rather than updating an existing HISTORY or COMMENT card:

>>> prihdr['history'] = 'I updated this file 2/26/09'
>>> prihdr['comment'] = 'Edwin Hubble really knew his stuff'
>>> prihdr['comment'] = 'I like using HST observations'
>>> prihdr['history']
I updated this file 2/26/09
>>> prihdr['comment']
Edwin Hubble really knew his stuff
I like using HST observations

Note: Be careful not to confuse COMMENT cards with the comment value for normal cards.

To updating existing COMMENT or HISTORY cards, reference them by index:

>>> prihdr['history'][0] = 'I updated this file on 2/27/09'
>>> prihdr['history']
I updated this file on 2/27/09
>>> prihdr['comment'][1] = 'I like using JWST observations'
>>> prihdr['comment']
Edwin Hubble really knew his stuff
I like using JWST observations

To see the entire header as it appears in the FITS file (with the END card and padding stripped), simply enter the header object by itself, or print repr(header):

>>> header
SIMPLE  =                    T / file does conform to FITS standard
BITPIX  =                   16 / number of bits per data pixel
NAXIS   =                    0 / number of data axes
...all cards are shown...
>>> print repr(header)

Entering simply print header will also work, but may not be very legible on most displays, as this displays the header as it is written in the FITS file itself, which means there are no linebreaks between cards. This is a common confusion in new users of PyFITS.

It’s also possible to view a slice of the header:

>>> header[:2]
SIMPLE  =                    T / file does conform to FITS standard
BITPIX  =                   16 / number of bits per data pixel

Only the first two cards are shown above.

To get a list of all keywords, use the Header.keys() method just as you would with a dict:

>>> prihdr.keys()
['SIMPLE', 'BITPIX', 'NAXIS', ...]

Working with Image Data

If an HDU’s data is an image, the data attribute of the HDU object will return a numpy ndarray object. Refer to the numpy documentation for details on manipulating these numerical arrays.

>>> scidata = hdulist[1].data

Here, scidata points to the data object in the second HDU (the first HDU, hdulist[0], being the primary HDU) which corresponds to the ‘SCI’ extension. Alternatively, you can access the extension by its extension name (specified in the EXTNAME keyword):

>>> scidata = hdulist['SCI'].data

If there is more than one extension with the same EXTNAME, the EXTVER value needs to be specified along with the EXTNAME as a tuple; e.g.:

>>> scidata = hdulist['sci', 2].data

Note that the EXTNAME is also case-insensitive.

The returned numpy object has many attributes and methods for a user to get information about the array; e.g.

>>> scidata.shape
(800, 800)

Since image data is a numpy object, we can slice it, view it, and perform mathematical operations on it. To see the pixel value at x=5, y=2:

>>> print scidata[1, 4]

Note that, like C (and unlike FORTRAN), Python is 0-indexed and the indices have the slowest axis first and fastest changing axis last; i.e. for a 2-D image, the fast axis (X-axis) which corresponds to the FITS NAXIS1 keyword, is the second index. Similarly, the 1-indexed sub-section of x=11 to 20 (inclusive) and y=31 to 40 (inclusive) would be given in Python as:

>>> scidata[30:40, 10:20]

To update the value of a pixel or a sub-section:

>>> scidata[30:40, 10:20] = scidata[1, 4] = 999

This example changes the values of both the pixel [1, 4] and the sub-section [30:40, 10:20] to the new value of 999. See the Numpy documentation for more details on Python-style array indexing and slicing.

The next example of array manipulation is to convert the image data from counts to flux:

>>> photflam = hdulist[1].header['photflam']
>>> exptime = prihdr['exptime']
>>> scidata *= photflam / exptime

Note that performing an operation like this on an entire image requires holding the entire image in memory. This example performs the multiplication in-place so that no copies are made, but the original image must first be able to fit in main memory. For most observations this should not be an issue on modern personal computers.

If at this point you want to preserve all the changes you made and write it to a new file, you can use the HDUList.writeto() method (see below).

Working With Table Data

If you are familiar with numpy recarray (record array) objects, you will find the table data is basically a record array with some extra properties. But familiarity with record arrays is not a prerequisite for this guide.

Like images, the data portion of a FITS table extension is in the .data attribute:

>>> hdulist ='table.fits')
>>> tbdata = hdulist[1].data # assuming the first extension is a table

To see the first row of the table:

>>> print tbdata[0]
(1, 'abc', 3.7000002861022949, 0)

Each row in the table is a FITS_record object which looks like a (Python) tuple containing elements of heterogeneous data types. In this example: an integer, a string, a floating point number, and a Boolean value. So the table data are just an array of such records. More commonly, a user is likely to access the data in a column-wise way. This is accomplished by using the field() method. To get the first column (or “field” in Numpy parlance–it is used here interchangeably with “column”) of the table, use:

>>> tbdata.field(0)
array([1, 2])

A numpy object with the data type of the specified field is returned.

Like header keywords, a column can be referred either by index, as above, or by name:

>>> tbdata.field('id')
array([1, 2])

When accessing a column by name, dict-like access is also possible (and even preferable):

>>> tbdata['id']
array([1, 2])

In most cases it is preferable to access columns by their name, as the column name is entirely independent of its physical order in the table. As with header keywords, column names are case-insensitive.

But how do we know what columns we have in a table? First, let’s introduce another attribute of the table HDU: the columns attribute:

>>> cols = hdulist[1].columns

This attribute is a ColDefs (column definitions) object. If we use the method from the interactive prompt:

      ['c1', 'c2', 'c3', 'c4']
      ['1J', '3A', '1E', '1L']
      ['', '', '', '']
      [-2147483647, '', '', '']
      ['', '', 3, '']
      ['', '', 0.40000000000000002, '']
      ['I11', 'A3', 'G15.7', 'L6']
      ['', '', '', '']
      ['', '', '', '']

it will show the attributes of all columns in the table, such as their names, formats, bscales, bzeros, etc. A similar output that will display the column names and their formats can be printed from within a script with:

print hdulist[1].columns

We can also get these properties individually; e.g.

>>> cols.names
['ID', 'name', 'mag', 'flag']

returns a (Python) list of field names.

Since each field is a numpy object, we’ll have the entire arsenal of Numpy tools to use. We can reassign (update) the values:

>>> tbdata.['flag'][:] = 0

take the mean of a column:

>>> tbdata['mag'].mean()
>>> 84.4

and so on.

Save File Changes

As mentioned earlier, after a user opened a file, made a few changes to either header or data, the user can use HDUList.writeto() to save the changes. This takes the version of headers and data in memory and writes them to a new FITS file on disk. Subsequent operations can be performed to the data in memory and written out to yet another different file, all without recopying the original data to (more) memory.

>>> hdulist.writeto('newimage.fits')

will write the current content of hdulist to a new disk file newfile.fits. If a file was opened with the update mode, the HDUList.flush() method can also be used to write all the changes made since open(), back to the original file. The close() method will do the same for a FITS file opened with update mode:

>>> f ='original.fits', mode='update')
... # making changes in data and/or header
>>> f.flush()  # changes are written back to original.fits
>>> f.close()  # closing the file will also flush any changes and prevent
...            # further writing

Creating a New FITS File

Creating a New Image File

So far we have demonstrated how to read and update an existing FITS file. But how about creating a new FITS file from scratch? Such task is very easy in PyFITS for an image HDU. We’ll first demonstrate how to create a FITS file consisting only the primary HDU with image data.

First, we create a numpy object for the data part:

>>> import numpy as np
>>> n = np.arange(100.0) # a simple sequence of floats from 0.0 to 99.9

Next, we create a PrimaryHDU object to encapsulate the data:

>>> hdu = pyfits.PrimaryHDU(n)

We then create a HDUList to contain the newly created primary HDU, and write to a new file:

>>> hdulist = pyfits.HDUList([hdu])
>>> hdulist.writeto('new.fits')

That’s it! In fact, PyFITS even provides a short cut for the last two lines to accomplish the same behavior:

>>> hdu.writeto('new.fits')

This will write a single HDU to a FITS file without having to manually encapsulate it in an HDUList object first.

Creating a New Table File

To create a table HDU is a little more involved than image HDU, because a table’s structure needs more information. First of all, tables can only be an extension HDU, not a primary. There are two kinds of FITS table extensions: ASCII and binary. We’ll use binary table examples here.

To create a table from scratch, we need to define columns first, by constructing the Column objects and their data. Suppose we have two columns, the first containing strings, and the second containing floating point numbers:

>>> import pyfits
>>> import numpy as np
>>> a1 = np.array(['NGC1001', 'NGC1002', 'NGC1003'])
>>> a2 = np.array([11.1, 12.3, 15.2])
>>> col1 = pyfits.Column(name='target', format='20A', array=a1)
>>> col2 = pyfits.Column(name='V_mag', format='E', array=a2)

Next, create a ColDefs (column-definitions) object for all columns:

>>> cols = pyfits.ColDefs([col1, col2])

Now, create a new binary table HDU object by using the PyFITS function BinTableHDU.from_columns():

>>> tbhdu = pyfits.BinTableHDU.from_columns(cols)

This function returns (in this case) a BinTableHDU.

Of course, you can do this more concisely without creating intermediate variables for the individual columns and without manually creating a ColDefs object:

>>> tbhdu = pyfits.BinTableHDU.from_columns([
...     pyfits.Column(name='target', format='20A', array=a1),
...     pyfits.Column(name='V_mag', format='E', array=a2)])

Now you may write this new table HDU directly to a FITS file like so:

>>> tbhdu.writeto('table.fits')

This shortcut will automatically create a minimal primary HDU with no data and prepend it to the table HDU to create a valid FITS file. If you require additional data or header keywords in the primary HDU you may still create a PrimaryHDU object and build up the FITS file manually using an HDUList.

For example, first create a new Header object to encapsulate any keywords you want to include in the primary HDU, then as before create a PrimaryHDU:

>>> prihdr = pyfits.Header()
>>> prihdr['OBSERVER'] = 'Edwin Hubble'
>>> prihdr['COMMENT'] = "Here's some commentary about this FITS file."
>>> prihdu = pyfits.PrimaryHDU(header=prihdr)

When we create a new primary HDU with a custom header as in the above example, this will automatically include any additional header keywords that are required by the FITS format (keywords such as SIMPLE and NAXIS for example). In general, PyFITS users should not have to manually manage such keywords, and should only create and modify observation-specific informational keywords.

We then create a HDUList containing both the primary HDU and the newly created table extension, and write to a new file:

>>> thdulist = pyfits.HDUList([prihdu, tbhdu])
>>> thdulist.writeto('table.fits')

Alternatively, we can append the table to the HDU list we already created in the image file section:

>>> hdulist.append(tbhdu)
>>> hdulist.writeto('image_and_table.fits')

The data structure used to represent FITS tables is called a FITS_rec and is derived from the numpy.recarray interface. When creating a new table HDU the individual column arrays will be assembled into a single FITS_rec array.

So far, we have covered the most basic features of PyFITS. In the following chapters we’ll show more advanced examples and explain options in each class and method.

Convenience Functions

PyFITS also provides several high level (“convenience”) functions. Such a convenience function is a “canned” operation to achieve one simple task. By using these “convenience” functions, a user does not have to worry about opening or closing a file, all the housekeeping is done implicitly.


These functions are useful for interactive Python sessions and simple analysis scripts, but should not be used for application code, as they are highly inefficient. For example, each call to getval() requires re-parsing the entire FITS file. Code that makes repeated use of these functions should instead open the file with open() and access the data structures directly.

The first of these functions is getheader(), to get the header of an HDU. Here are several examples of getting the header. Only the file name is required for this function. The rest of the arguments are optional and flexible to specify which HDU the user wants to access:

>>> from pyfits import getheader
>>> getheader('in.fits')  # get default HDU (=0), i.e. primary HDU's header
>>> getheader('in.fits', 0)  # get primary HDU's header
>>> getheader('in.fits', 2)  # the second extension
>>> getheader('in.fits', 'sci')  # the first HDU with EXTNAME='SCI'
>>> getheader('in.fits', 'sci', 2)  # HDU with EXTNAME='SCI' and EXTVER=2
>>> getheader('in.fits', ('sci', 2))  # use a tuple to do the same
>>> getheader('in.fits', ext=2)  # the second extension
>>> getheader('in.fits', extname='sci')  # first HDU with EXTNAME='SCI'
>>> getheader('in.fits', extname='sci', extver=2)

Ambiguous specifications will raise an exception:

>>> getheader('in.fits', ext=('sci', 1), extname='err', extver=2)
TypeError: Redundant/conflicting extension arguments(s): {'ext': ('sci',
1), 'args': (), 'extver': 2, 'extname': 'err'}

After you get the header, you can access the information in it, such as getting and modifying a keyword value:

>>> from pyfits import getheader
>>> hdr = getheader('in.fits', 1)  # get first extension's header
>>> filter = hdr['filter']         # get the value of the keyword "filter'
>>> val = hdr[10]                  # get the 11th keyword's value
>>> hdr['filter'] = 'FW555'        # change the keyword value

For the header keywords, the header is like a dictionary, as well as a list. The user can access the keywords either by name or by numeric index, as explained earlier in this chapter.

If a user only needs to read one keyword, the getval() function can further simplify to just one call, instead of two as shown in the above examples:

>>> from pyfits import getval
>>> flt = getval('in.fits', 'filter', 1)  # get 1st extension's FILTER
>>> val = getval('in.fits', 10, 'sci', 2)  # get the value 2nd SCI
...                                        # extension's 11th keyword

The function getdata() gets the data of an HDU. Similar to getheader(), it only requires the input FITS file name while the extension is specified through the optional arguments. It does have one extra optional argument header. If header is set to True, this function will return both data and header, otherwise only data is returned:

>>> from pyfits import getdata
>>> dat = getdata('in.fits', 'sci', 3)  # get 3rd sci extension's data
... # get 1st extension's data and header
>>> data, hdr = getdata('in.fits', 1, header=True)

The functions introduced above are for reading. The next few functions demonstrate convenience functions for writing:

>>> pyfits.writeto('out.fits', data, header)

The writeto() function uses the provided data and an optional header to write to an output FITS file.

>>> pyfits.append('out.fits', data, header)

The append() function will use the provided data and the optional header to append to an existing FITS file. If the specified output file does not exist, it will create one.

>>> from pyfits import update
>>> update(file, dat, hdr, 'sci')        # update the 'sci' extension
>>> update(file, dat, 3)                 # update the 3rd extension
>>> update(file, dat, hdr, 3)            # update the 3rd extension
>>> update(file, dat, 'sci', 2)          # update the 2nd SCI extension
>>> update(file, dat, 3, header=hdr)     # update the 3rd extension
>>> update(file, dat, header=hdr, ext=5) # update the 5th extension

The update() function will update the specified extension with the input data/header. The 3rd argument can be the header associated with the data. If the 3rd argument is not a header, it (and other positional arguments) are assumed to be the extension specification(s). Header and extension specs can also be keyword arguments.

Finally, the info() function will print out information of the specified FITS file:

Filename: test0.fits
No. Name    Type       Cards Dimensions Format
0   PRIMARY PrimaryHDU   138 ()         Int16
1   SCI     ImageHDU      61 (400, 400) Int16
2   SCI     ImageHDU      61 (400, 400) Int16
3   SCI     ImageHDU      61 (400, 400) Int16
4   SCI     ImageHDU      61 (400, 400) Int16

This is one of the most useful convenience functions for getting an overview of what a given file contains without looking at any of the details.



A module for reading and writing FITS files and manipulating their contents.

A module for reading and writing Flexible Image Transport System (FITS) files. This file format was endorsed by the International Astronomical Union in 1999 and mandated by NASA as the standard format for storing high energy astrophysics data. For details of the FITS standard, see the NASA/Science Office of Standards and Technology publication, NOST 100-2.0.

For detailed examples of usage, see the PyFITS User’s Manual.