This is an archival dump of old wiki content --- see for current material

Building on Windows

From Gary Pajer, post to scipy-dev, 20050104 rev 20050105:

The svn versions of numpy and scipy build extremely smoothly on WinXP using the MinGW compiler. It's a lot easier than these directions make it sound. To build numpy, nothing else is needed. To build scipy, you also need the ATLAS library.

Step 0: get ready

numpy and scipy build and install separately. You must first build and install numpy, then do scipy. Before you can build, you will have to create two directories to act as workspaces, one for numpy, the other for scipy.

Step 1: Getting MinGW

The hardest part of building numpy/scipy is finding the right thing to download from MingGW.

Start here

Click on "Download MinGW" (in the green box)

Click on "Previous" (non intuitive!)

Click on "MingGW" (down the page a bit)

Download MinGW-3.1.0-1.exe <> (version 4.1.0 doesn't work for me)

and install

Step 2: numpy

Here's an updated version of instructions that Travis posted a while ago:

Check out the latest numpy SVN tree into your numpy workspace directory (Tortoise SVN is an excellent windows SVN client that makes it easy). The URL is

You should be able to go into the directory where you placed the tree and type

python config --compiler=mingw32 build --compiler=mingw32 install


python config --compiler=mingw32 build --compiler=mingw32 bdist_wininst

to get an installable executable.

Alternatively, to avoid all the --compiler=xxxxx noise you can create (or modify if you already have one) a distutils configuration file for your version of Python. The file name is

<your python directory>\Lib\distutils\distutils.cfg

and the contents should contain

compiler = mingw32

compiler = mingw32

On my system


is where it is located.

Step 3: scipy

You need an ATLAS library to build scipy. For an easy build, use a prebuilt binary from here: and install. set an environment variable set ATLAS=c:\path\to\atlas (or set via the Windows control panel) Of course, you can build ATLAS yourself if you want to.

Then just get yourself into your scipy workspace directory, check out the SVN from and then build using the same command that built numpy

Step 4: Do science

SciPy: Build_for_Windows (last edited 2015-10-24 17:48:25 by anonymous)