The Gene Ontology™ (GO) of biological process, molecular function and cellular component terms is the predominant source for functional annotation of gene products. An important use of GO-based annotation has been in the interpretation of gene expression microarray results. One of the challenges to gene expression microarray data analysis and interpretation is that cross-hybridization of probes to the related transcripts can contribute to the signal measured. Several recent studies have reported revised microarray probe annotations designed to circumvent this problem by ensuring that the probe annotation matches the current version of the relevant genome sequence and by eliminating probes with sequence similarity to multiple gene, but the impact of these revised annotations remains to be assessed. Here we describe a general approach of using GO annotation coclustering characteristics to compare the performance of alternative data mining methods, and apply this approach to assess the impact of improved probe annotation on the results of gene expression microarray data interpretation. Using this approach, we found that revised Affymetrix GeneChip® probe annotation gives rise to improved interpretation of microarray gene expression experiments related to the development, function and transformation of human B lymphocytes.