The potential use of nude mouse xenografts as a source of human tumor tissue for preclinical assessment of drug activity was examined by a comparison of in vitro sensitivity to four anthracycline derivatives of eight xenografts maintained in nude mice and tumor colony-forming units (CFU) from the xenografts grown in agar. Results obtained in the two systems with doxorubicin (Dx) and a new, closely related derivative (epi-doxorubicin, epi-Dx) correlated well. In vivo tumor growth delay and maximum in vitro tumor CFU kill for these two drugs showed a significant correlation (r = 0.64, P less than .01). Separation of tumors into "sensitive" and "insensitive" tumor populations on the basis of maximum in vitro CFU kill also predicted the in vivo response to these two drugs in 88% of cases. Similar analyses performed on in vitro and in vivo results with two other new anthracyclines (4'-deoxy-doxorubicin, deoxy-Dx and 4'-O-methyl-doxorubicin, O-Me-Dx) showed a significant negative correlation between in vivo and in vitro results; the in vitro system failed to predict in vivo activity of these two drugs. No significant differences in in vitro activity against normal, granulocyte/macrophage progenitors (CFU-GM) or against various tumor CFUs were detected. Thus, the selectivity (activity against normal tissue compared with that against tumor) of the four drugs appeared equal. These data suggest that in vitro screening of drugs using tumor CFUs from nude mouse xenografts may predict the in vivo activity of drugs for which pharmacologic data are available, but illustrate the difficulties in attempting to predict the in vivo activity of new drugs for which no such data are available.