Lymphatic mapping and sentinel node biopsy are well-established techniques for staging and managing patients with melanoma, breast cancer and other malignancies that spread initially to the regional lymph nodes. Identification of tumor in the sentinel node is the most precise staging technique currently available. The sentinel node is the site of metastatic melanoma in approximately 20% of melanoma patients and if tumor is present in the sentinel node it is customary to perform a complete dissection of the lymph nodes of the affected nodal basin. This may be overtreatment for some patients as tumor is identified in the nonsentinel nodes of only one-third of sentinel node-positive melanoma patients treated by completion lymphadenectomy. If it were possible accurately to identify the minority of patients with tumor in the nonsentinel nodes, the patients most likely to benefit from lymphadenectomy, the remaining patients could be spared a potentially morbid operation that is unlikely to confer clinical advantage. In 90 patients with a melanoma-positive sentinel node, who subsequently had a completion lymphadenectomy, we evaluated and compared the capacity of characteristics of the primary melanoma and of the sentinel node to predict individuals likely to have tumor in nonsentinel nodes. We assessed the Breslow thickness of the primary, the amount of tumor in the sentinel node (relative tumor area) and, as an index of immune modulation of the sentinel node, the density of dendritic leukocytes in the nodal paracortex. The relative area of tumor in the sentinel node and Breslow thickness of the primary melanoma most accurately predicted the presence of tumor in the nonsentinel nodes (P=0.0001 in both cases-Wilcoxon rank sums). The presence of melanoma in the nonsentinel nodes was also predicted by the density of dendritic leukocytes in the paracortex (P=0.008-Wilcoxon rank sums). These three observations assessed alone and in combination predict the presence of tumor in the nonsentinel nodes with high accuracy. The same characteristics also significantly correlated with tumor recurrence (tumor burden, P=0.0001, Breslow, P=0.0001 and dendritic cell density, P=0.0007) and death from melanoma (tumor burden, P=0.0001, Breslow, P=0.0001 and dendritic cell density, P=0.0026).