Sentinel lymph node biopsy is increasingly used to identify occult metastases in regional lymph nodes of patients with melanoma. Selection of patients for sentinel lymph node biopsy and subsequent lymphadenectomy is an area of debate. The purpose of this study was to describe a large clinical series of these biopsies for cutaneous melanoma and to identify patients most likely to gain useful clinical information from sentinel lymph node biopsy. The Indiana University Melanoma Program computerized database was queried to identify all patients who underwent this procedure for clinically localized cutaneous melanoma. It was performed using preoperative technetium Tc 99m lymphoscintigraphy and isosulfan blue dye. Pertinent demographic, surgical, and histopathologic data were recorded. Univariate and multivariate logistic regression and classification table analyses were performed to identify clinical variables associated with sentinel node and nonsentinel node positivity. In total, 234 biopsy procedures were performed to stage 291 nonpalpable regional lymph node basins. Mean Breslow's thickness was 2.30 mm (2.08 mm for negative sentinel lymph node biopsy, 3.18 mm for positive). The mean number of sentinel nodes removed was 2.17 nodes per basin (range, 1 to 8). Forty-seven of 234 melanomas (20.1 percent) and 50 of 291 basins (17.2 percent) had a positive biopsy. Positivity correlated with AJCC tumor stage: T1, 3.6 percent; T2, 8.1 percent; T3, 27.4 percent; T4, 44 percent. By univariate logistic regression, Breslow's thickness (p = 0.003, continuous variable), ulceration (p = 0.003), mitotic index > or = 6 mitoses per high power field (p = 0.008), and Clark's level (p = 0.04) were significantly associated with sentinel lymph node biopsy result. By multivariate analysis, only Breslow's thickness (p = 0.02), tumor ulceration (p = 0.02), and mitotic index (p = 0.02) were significant predictors of biopsy positivity. Classification table analysis showed the Breslow cutpoint of 1.2 mm to be the most efficient cutpoint for sentinel lymph node biopsy result (p = 0.0004). Completion lymphadenectomy was performed in 46 sentinel node-positive patients; 12 (26.1 percent) had at least one additional positive nonsentinel node. Nonsentinel node positivity was marginally associated with the presence of multiple positive sentinel nodes (p = 0.07). At mean follow-up of 13.8 months, four of 241 sentinel node-negative basins demonstrated same-basin recurrence (1.7 percent). Sentinel lymph node biopsy is highly reliable in experienced hands but is a low-yield procedure in most thin melanomas. Patients with melanomas thicker than 1.2 mm or with ulcerated or high mitotic index lesions are most likely to have occult lymph node metastases by sentinel lymph node biopsy. Completion therapeutic lymphadenectomy is recommended after positive biopsy because it is difficult to predict the presence of positive nonsentinel nodes.