To evaluate the ability of high-resolution ultrasonography (hrUS) to detect sentinel-node (SN) melanoma metastases preoperatively before sentinel-node biopsy (SNB), to define hrUS resolution, and to evaluate which US criteria should be used. During a 6.5-year period, 131 consecutive patients with 132 >or=1-mm thick or ulcerated cutaneous melanomas, who were followed up at a single center, were enrolled. All patients underwent preoperative regional lymph-node hrUS and SNB. We used the recently evaluated ultrasonographic stringent and nonstringent hrUS criteria to detect SN metastases. Sizes of the SN metastatic deposits were measured under light microscopy. Thirty-five (27%) patients had a positive SNB. HrUS identified only three positive SNs as being metastatic. Sensitivity and specificity using stringent criteria were 8.8% [95% confidence interval (CI, 2.3-24.8%) and 95.9% (95% CI, 89.3-98.7%)], respectively. Positive-predictive value was 42.9% (95% CI, 11.9-79.9%). The nonstringent criteria provided four additional true-positive results, but lowered specificity (89.8%; 95% CI, 81.6-94.7%) with no significant improvement in sensitivity (20.6%; 95% CI, 9.3-38.4%). Positive-predictive value using nonstringent criteria was 41.2% (95% CI, 19.3-66.4%). HrUS failed to detect all metastatic deposits <5 mm in diameter. HrUS assessment of early-stage melanomas cannot replace surgical SNB. Owing to its low positive-predictive value, hrUS was unable to identify patients who would have to proceed directly to completion lymphadenectomy.