The purpose of these studies was to select and isolate cells with increased liver-metastasizing potential from heterogeneous primary human colon carcinomas (HCCs). Cells derived from a primary HCC classified as Dukes' stage B2 were directly established in culture or were injected into the subcutis, cecum, or spleen of nude mice. Progressively growing tumors were excised, dissociated, and established in culture. Subsequent to implantation into the cecum or spleen of nude mice, cells from all four lines produced only a few liver tumor foci. HCC cells from the few liver metastases were expanded in culture and then injected into the spleen of nude mice to provide a source for further cycles of selection. With each successive in vivo selection cycle, the metastatic ability of the isolated propagated cells increased. Four cycles of selection yielded cell lines with a very high metastatic efficiency in nude mice. In parallel studies using another primary HCC classified as Dukes' stage D, we isolated cell lines that were highly metastatic in nude mice. Successive selection cycles for growth in the liver increased the metastatic properties of the HCC cells, albeit to a lesser extent than it did those of the Dukes' B2 stage HCC. The ability of the HCC cells to produce liver metastases was not due to simple trapping in the liver. In vivo distribution studies using [125I] iododeoxyuridine-labeled tumor cells revealed that, shortly after injection into the spleen, a comparable number of cells with either low or high metastatic properties arrested in the liver. The differences between the low- and high-degree metastatic cells became apparent by 24 h after injection and, by 72 h, only highly metastatic cells survived in the liver. These results demonstrate that hepatic metastasis by HCC cells is a selective process and that the nude mouse model can be useful for isolating highly metastatic HCC cells and for studying the relevant host organ factors that regulate the pathogenesis of metastasis.