To study the usefulness of an in vitro colony-forming assay in predicting individual clinical responses to chemotherapy, tumor cells obtained from 150 melanoma metastases (119 patients) were grown in soft agar according to the method of Courtenay and Mills (1978), and tested for sensitivity to DTIC, CCNU, vinblastine, procarbazine, abrin and ricin. In 83% of the cases colony formation was observed (plating efficiency, PE, greater than 0.01%). Twenty-seven per cent of the tumors gave PEs greater than 1%, 45% gave PEs in the range 0.1-0.9%, whereas 11% of the tumors gave 0.01-0.09%. The PEs were not correlated with the degree of pigmentation or with the clinical course. Evaluable chemosensitivity data were obtained on 104 metastases from 83 patients. Large differences in sensitivity were seen. In cases which were evaluable both in vivo and in vitro a clear correlation was found between the in vitro chemosensitivity, expressed as the expected growth delay, and the clinical response to chemotherapy. Tumors from patients with partial response, mixed response or stable disease after prior progression, all had rather high in vitro sensitivity to the drug used (expected growth delay greater than 2.0), whereas patients with progression had lower sensitivity. The results confirm that the soft agar method used here provides good culture conditions for human melanoma cells and show that chemosensitivity data can be obtained in a high percentage of melanoma patients. The approach used seems promising in aiding clinicians to adjust chemotherapy to individual patients.