In the September 11 Proceedings of the National Academy of Sciences, Sorlie et al. show how cDNA microarray data may be used to define subclasses of breast carcinomas and to predict clinical outcome (Proc Natl Acad Sci USA 2001, 98:10869-10874). They analysed 78 breast cancer samples (mostly ductal carcinomas) and compared the expression profiles of 456 genes with profiles from normal breast tissue. They then classified the cancers into epithelium-like, ERBB2-overexpressing and normal-breast-like groups, and used different profiling patterns to sub-divide the estrogen-receptor-positive group into two distinct subgroups. Sorlie et al. found that the microarray-defined subgroups of patients had different clinical outcomes, suggesting the power of microarray-based studies for determining prognoses.