Chemotherapy for breast cancer is given on the basis of empirical information from clinical trials, an approach which falls to take into account the known heterogeneity of chemosensitivity between patients. Previous attempts to determine chemosensitivity ex vivo have been disappointing, but in this study results from a newly developed tumor chemosensitivity assay (TCA) have been correlated prospectively with patient response. In this study, we have used heterogeneity data for standard regimens obtained from 116 breast TCAs to set sensitivity/resistance thresholds which were then used to interpret the results from those with known clinical responses. Assay evaluability was 97% in surgical biopsies. Clinical follow-up of stage III/ IV assessable disease was obtained from 27 breast tumors which were successfully tested for chemosensitivity, including 13 needle biopsies. The ATP-TCA assay predicted response correctly in 22 out of 29 (76%) tumors with clinically evaluable disease, suggesting that it is capable of predicting outcome in individual patients. Assays were performed in seven patients before and after chemotherapy using residual or recurrent tumor tissue. Four cases with initial sensitivity showed a decrease in sensitivity within 6 months of starting chemotherapy, while two others without clinical resistance were still sensitive by TCA. All nine courses of therapy given on the basis of TCA sensitivity resulted in partial or complete responses. Controlled trials of TCA-directed treatment against standardized empirical therapy should be conducted before this technology is widely adopted to assess its impact on rates of response, survival and the cost of treatment.