I have used Matlab to a limited extent, primarily to use others' algorithms and modify their code, but amongst computer scientists who do statistics and machine learning, it is very popular. Matlab is a scripting language and the syntax and feel of it are similar to R and S. Tim Hoar at NCAR swears by it and has a nice introduction. My understanding is that one of the really nice things about it is that the graphics are highly customizable - you can create a graphic and then get an object that has all the attributes of the graphic that you can change. So if you're not able to customize a graph sufficiently in R or Splus, such as for a manuscript figure, you might want to look into doing it in Matlab. I've seen some claims that Matlab is generally faster than R, but limited experience suggests this is not always true, although when I recently compared Cholesky decompositions in Matlab 6.5 and R 1.7.1 of 400 by 400 and 1000 by 1000 matrices, Matlab was almost twice as fast. To my mind, the major downside of Matlab is that it is commercial software (though there is an open-source version called octave). Also, while it has many statistical features (and more may be available in various toolboxes), it still requires writing more of your own code than R or S.

- Converting between R or S and Matlab
- Command line operation (no GUI)
- Loading and saving large objects/files
- Run-time calculations