Correlated variables have been shown to confound statistical analyses in microarray experiments. The same effect applies to an even greater degree in proteomics, especially with the use of MS for parallel measurements. Biological effects such as PTM, fragmentation, and multimer formation can produce strongly correlated variables. The problem is compounded in some types of MS by technical effects such as incomplete chromatographic separation, binding to multiple surfaces, or multiple ionizations. Existing methods for dimension reduction, notably principal components analysis and related techniques, are not always satisfactory because they produce data that often lack clear biological interpretation. We propose a preprocessing algorithm that clusters highly correlated features, using the Bayes information criterion to select an optimal number of clusters. Statistical analysis of clusters, instead of individual features, benefits from lower noise, and reduces the difficulties associated with strongly correlated data. This preprocessing increases the statistical power of analyses using false discovery rate on simulated data. Strong correlations are often present in real data, and we find that clustering improves biomarker discovery in clinical SELDI-TOF-MS datasets of plasma from patients with Kawasaki disease, and bone-marrow cell extracts from patients with acute myeloid or acute lymphoblastic leukemia.