An efficient node-deletion algorithm is introduced to find submatrices in expression data that have low mean squared residue scores and it is shown to perform well in finding co-regulation patterns in yeast and human. This introduces "biclustering", or simultaneous clustering of both genes and conditions, to knowledge discovery from expression data. This approach overcomes some problems associated with traditional clustering methods, by allowing automatic discovery of similarity based on a subset of attributes, simultaneous clustering of genes and conditions, and overlapped grouping that provides a better representation for genes with multiple functions or regulated by many factors.