Genomic expression signatures provide high-content biomarkers of cellular physiology, including the diverse responses to therapeutic drugs. To recognize these signatures, we devised a method of biomarker evaluation called 'sampling over gene space' (SOGS) that imparts superior predictive performance to existing supervised classification algorithms. Applied to microarray data from drug-treated human cortical neuron 1A cell cultures, this method predicts whether individual compounds possess anticonvulsant, antihypertensive, cyclooxygenase inhibitor, or opioid action. Thus, stable cell lines can be suitable for expression signature-based screening of a diverse range of activities. A SOGS-based system also discriminates physiologically active from inactive compounds, identifies drugs with off-target side effects, and incorporates a quantitative method for assigning confidence to individual predictions that, at its most stringent, approaches 100% accuracy. The capacity to resolve multiple distinct drug activities while simultaneously discriminating inactive and potential false-positive compounds in a cell line presents a unified framework for streamlined chemical genomic drug discovery.