Identification of the cellular and subcellular alterations responsible for the metastatic behavior of malignant tumor cells and development of reliable screening programs for detecting new therapeutic agents for improved treatment of metastatic disease both depend crucially on the availability of experimental systems that can serve as relevant models of human cancer. Recent advances in our understanding of the pathogenesis of cancer metastasis have raised serious doubts about the usefulness of many of the experimental approaches that have long been used in the study of metastasis. Recent findings showing that metastases are caused by specific subpopulations of metastatic tumor cells, and that not all cells in a malignant primary tumor possess metastatic properties, are of profound importance for experimental efforts to understand the mechanism of metastatic phenotype among cells from the same tumor means that the traditional, and widely used, approach of analyzing primary tumors and cultured cell lines containing multiple, phenotypically heterogeneous, subpopulations of cells may provide little or no insight into the properties of the metastatic subpopulations, particularly if they represent only a minor fraction of the entire population. Similarly, the practice of screening potential therapeutic modalities for their ability to reduce the mass and/or growth rate of a primary tumor may be inadequate in predicting the responsiveness of metastatic lesions. Solution of these problems requires that new methods must be devised to isolate and characterize the specific subpopulations of tumor cells endowed with metastatic potential. In addition, knowledge of how the extraordinary phenotypic diversity found in tumor cell subpopulations from the same tumor is generated and how subpopulation diversity is regulated during progressive growth of both the primary tumor and its metastases are of fundamental importance if we are to design meaningful experimental systems for studying the metastatic process. This article reviews our current understanding of these complex issues and their implications for the experimental analysis of the malignant phenotype. The merits and shortcomings of different experimental systems are discussed in detail together with the identification of areas in which new experimental strategies and models are now needed.'