Application of gene expression technology to cancer pharmacology and drug discovery is particularly challenging because most clinical tumour specimens have complex and imperfect treatment histories. In contrast, the 60 cell lines of the National Cancer Institute's drug discovery program can be likened to 60 patients in a clinical trial in which each patient has been treated with more than 70,000 different chemical compounds one at a time and independently. This analogy provided a major reason to characterize gene expression in the 60 cell lines, despite the obvious fact that cell lines are not fully representative of clinical tumours. Accordingly, we have used 9,704-gene cDNA microarrays to assess gene expression patterns in the cell lines and found correlations of those patterns with profiles of sensitivity to a set of 1,400 potential anticancer agents. In tandem with the experimental studies, we have developed web-based analytical and data visualization tools (http://discover.nci.nih.gov and http://rana.stanford.edu) for examination of the large amounts of data derived from the gene expression patterns, drug sensitivity profiles and molecular structure descriptors of the tested compounds.