Todd Miller has added a matplotlib.numerix module to allow matplotlib
to choose between Numeric or numarry.  See the header of that file for
information on how to choose between Numeric or Numarray from the
command line or using environment variables.  

For the most part this is seamless and should provide any problems.
Below is a status report of known issues

* divide array by float - Many of the matplotlib examples do things
  like exp(-t/2.0) where t is an array.  If you have 'from __future__
  import division (as matplotlib.matlab does) then you will get an
  error along the lines of 

    TypeError: unsupported operand type(s) for /: 'NumArray' and 'float'"

  Solution: use numarray 0.9 or later; for older versions, use
  divide(-t, 2.0)

* stock demo does not run with "TypeError: unsubscriptable object"

  Solution: array resize/reshape bug fixed in numarray CVS

* Use of convolve in csd demo fails with "ValueError: Invalid
  convolution mode"

  Solution: fixed in numarray CVS