Three model systems involving LOX human malignant melanoma cells in nude rats were used to compare the chemosensitivity of tumors growing in different tissues. Groups of 4-18 rats with either s.c. xenografts, lung tumor colonies, or bone metastases were treated with cisplatin, doxorubicin, dacarbazine, or mitozolomide. The antitumor effect in the s.c. model was expressed as specific growth delay, and in the experimental metastasis studies as relative increase in life span (RILS), calculated on the basis of observed disease-free survival. Cisplatin had a moderate but significant effect on the progression of LOX tumor growth in all three systems. Doxorubicin was clearly more efficacious, but for both drugs tumor-free survivors were rare or absent. Importantly, for each of the compounds the levels of response were roughly the same in all three models, with specific growth delay and RILS values in the range of 0.2-0.3 for cisplatin and 0.5-0.9 for doxorubicin. In contrast, a significant site-dependent difference in sensitivity of the LOX tumors was observed for two alkylating agents. Thus, dacarbazine, which temporarily caused complete regression of s.c. xenografts (specific growth delay = 21.0), showed a moderate activity in the lung tumor model (RILS = 1.0) but had only a limited effect (RILS = 0.4) on bone metastases. Mitozolomide gave a curative effect in 6 of 10 animals with s.c. and in 4 of 4 animals with lung tumors, whereas in the bone metastasis model it was only slightly superior to doxorubicin (RILS = 1.1). In preliminary attempts to elucidate the underlying mechanisms, no site-dependent differences in drug distribution and in two cellular detoxifying systems were detected. The data demonstrate the usefulness of the LOX models for studying the clinically relevant problem of site-dependent tumor response to chemotherapy.