We developed a reliable new model system for assaying liver metastasis using NOD/SCID/gamma(c)(null) (NOG) mice. Seven human pancreatic cancer cell lines were examined for their ability to form diverse metastatic foci in the livers of NOD/SCID and NOG mice. Capan-2 and PL45 showed no metastasis when seeded at up to 10(5) cells in both strains, and no BxPC-3 metastasis was observed in NOD/SCID mice. The NOD/SCID mouse model detected liver metastasis only in the AsPC-1 cell line when inoculated with >10(3) cells. In contrast, when inoculated with only 10(2) MIA PaCa-2, AsPC-1 and PANC-1 cells, liver metastasis was evident in 71.4% (5/7), 57.1% (4/7) and 37.5% (3/8) of the NOG mice, respectively. Capan-1 and BxPC-3 cells metastasized when seeded at 10(3) cells in 50% (5/10) and in 12.5% (1/8) of the mice, respectively. Using the NOG mouse model system, we established a highly metastatic cell line, liver metastasized-BxPC-3 (LM-BxPC-3), from liver metastatic foci formed by the relatively poorly metastatic parental BxPC-3 cell line. The gene expression profiles of parental and LM-BxPC-3 cells were compared, and we identified forty-five genes that were either upregulated or downregulated >4-fold in the LM-BxPC-3 cell line. We validated 9 candidate protein-coding sequences, and examined the correlation between their expression pattern and the in vivo liver metastatic potential of all 7 pancreatic cancer cell lines. Only S100A4 expression correlated with the ability to form liver metastases, as evaluated in our quantitative model of metastasis in NOG mice. These results suggested that S100A4 is a key regulator of liver metastasis in pancreatic cancer, and demonstrated the feasibility of using the quantitative metastasis model to search for and develop new anti-cancer therapies and novel drugs against this and other key molecules.