Random Matrix Theory is an interesting line of work which basically states that to the extent a matrix resembles a random matrix as identified by the distribution of the eigen values then that set of eigen values most likely represent random noise. This technique modifies the noisy eigen values and reconstitutes the correlation matrix.
Laloux, et al is one paper which describes the RMT Model. The SciLab script, rmt2.sci, contains a very simple demonstration of RMT. A java implementation is now included in akutan.