The locations of distant secondary tumors in many clinical cancers and animal tumors are nonrandom, and their distributions cannot be explained by simple anatomical or mechanical hypotheses based on the simple lodgment or trapping of tumor cell emboli in the first capillary bed encountered. Evidence from certain experimental tumor systems supports Paget's 'seed and soil' hypothesis on the nonrandom distributions of metastases, in which the unique properties of particular tumor cells ('seeds') and the different characteristics of each organ microenvironment ('soil') collectively determine the organ preference of metastasis. Experimentally, differential tumor cell adhesion to organ-derived microvessel endothelial cells and organ parenchymal cells, differential invasion of basement membranes and organ tissues, and differential responses to organ-derived growth-stimulatory and -inhibitory factors all appear to be important determinants in explaining the organ preference of metastasis. Each tumor system may achieve organ specificity because of its own unique set of multiple metastasis-associated properties and responses to host microenvironments. As neoplasms progress to more highly malignant states multisite metastases are more likely and organ-specific metastases may be masked or circumvented owing to stochastic events, tumor cell diversification, host selection processes, and increased production of tumor autocrine molecules that may modulate adhesion, invasion, growth, and other properties important in metastasis. The importance of each of these properties, however, appears to vary considerably among different metastatic tumor systems. These and other tumor cell and host properties may eventually be used to predict and explain the unique metastatic distributions of certain human malignancies.