Fifteen variables were tested for their value in predicting recurrent disease in 46 clinical Stage I melanoma patients with metastases to regional nodes. A stepwise proportional hazards general linear model (Cox multivariate analysis) separated these melanoma patients with regional node metastases into at least two risk groups. Twenty patients in the relatively low-risk group had a five-year disease-free survival of 80% (in spite of having nodal metastases). This compares to a five-year disease-free survival of 17.5% for 26 patients in the high-risk group (P less than 0.001, Lee-Desu Statistic). Criteria for the high-risk group required that a patient have only one of the following two values: (1) The number of regional lymph nodes that contained tumor divided by the total number of nodes removed x 100% (percentage of positive nodes) greater than or equal to 20%; or (2) a primary tumor thickness of greater than 3.5 mm (regardless of node percentage). Conversely, patients in the low-risk group had neither of the above features. The high-risk group could further be stratified by the lymphocytic response at the base of the tumor. These findings have direct immediate application to the elective regional node dissection controversy and to adjuvant therapy studies containing these patients.