It has been suggested that more accurate and less subjective methods would improve the classification of human breast tumors . Global gene expression profiling is widely used to examine the expression of thousands of genes in biological samples . Indeed, this technology has been used extensively in numerous breast cancer studies to: examine the effects of various therapies on gene transcripts ; identify differences in gene expression among different tumor tissues ; molecularly classify tumors ; and to predict prognosis and treatment outcomes . Attempts to use gene expression profiles to identify the ER, PR and ERBB2 status of human breast tumors have also been reported . A single probe set representative of each gene was informative to establish ER, PR and ERBB2 expression in breast tumor samples. However, we wondered whether the specificity and/or sensitivity of this method could be improved by using probe sets representative of multiple genes whose expression correlated with that of the hormone receptors and ERBB2. Many peer-reviewed journals require authors to deposit microarray data in public PD325901 depositories, such as the Gene Expression Omnibus or ArrayExpress , thereby making them publicly available for various applications . However, clinical information such as hormone receptor or ERBB2 status of breast tumor samples is not invariably provided with their global gene expression profiles. Knowledge of hormone receptor and ERBB2 status as well as the global gene expression profiles of breast tumor samples may permit more accurate prognostic tests to be developed and would strengthen the value of the many breast tumor gene expression profiles in public depositories. Here we used 8 independent datasets containing human breast tumor samples profiled on Affymetrix Dinaciclib CDK inhibitor GeneChips to define gene expression signatures predictive of their ER and PR status as well as that of ERBB2. These gene signatures reliably predicted the status of the hormone receptors and that of ERBB2 as assessed by protein or DNA based tests. Because the largest predictive signature defined in our study comprises only 51 genes, a qRT-PCR based format may be developed that could provide an objective and relatively high-throughput alternative for the IHCbased definitions of hormone receptor and ERBB2 status in patient samples.