I highly recommend getting all of the components of Scipy-stack, which is just below the scipy files linked in the comment above. Unfortunately the maintainer hasn't updated scipy-stack to 3.3 yet (I emailed him about it), but all of the components (numpy-MKL, scipy, matplotlib, ipython, pandas, sympy, and nose), as well as all the dependencies (Python-dateutil, distribute, gmpy, PIL, pygments, pyreadline, pytz, statsmodels, and tornado) have 3.3 versions available for both Win32 and AMD64. Depending on what sort of computing you'll be doing, these packages will give you a great start, and as they're all from the same source they should all work well together.
UPDATE
I just heard from Christoph Gohlke, and he is holding off on updating scipy-stack until numpy-1.7.1, scipy-0.12, and ipython-0.13.2 are released, so probably a few months. If you want to build your own version, his redist_wininst.py script lets you do just that.
UPDATE 2
scipy-stack has been updated with new versions of its components, for both Python 2.7 and 3.3. Enjoy!
