We are actively developing visualization and analysis techniques to rapidly extract trends in composition-structure-property relationships from large data sets from combinatorial libraries. Because data often take on spectral or higher dimensional formats, it is helpful to apply data reduction schemes involving cluster analysis. Examples of techniques we have applied to date include hierarchical clustering using metric multidimensional scaling as the metric, non-negative matrix factorization, and the mean shift theory. They have proven to be extremely useful in deciphering the distribution of structural phases across composition spread samples using diffraction data. We have also applied our techniques to Raman spectra and hysteresis curves. We will also discuss applications of new regression techniques including the relevant vector machines. This work is funded by ONR and DOE.