The optimal surveillance for patients with resected high-risk melanoma is controversial. Select locoregional or oligometastatic recurrences can be cured with salvage resection. Data on the ability of PET/CT to detect such recurrences are sparse. We evaluated whether surveillance PET/CT in patients with resected stage III-IV melanoma led to detection of clinically occult recurrences amenable to curative-intent salvage treatment. We retrospectively identified 1429 melanoma patients who underwent PET/CT between January 2008 and October 2012 at Mayo Clinic (Rochester, Minnesota). A total of 1130 were excluded because of stage I-II, ocular or mucosal melanoma, incomplete resection, PET/CT not performed for surveillance or performed at a different institution, and records not available. A total of 299 patients were eligible. Overall, 162 (52%) patients developed recurrence [locoregional: 77 (48%), distant: 85 (52%)]. The first recurrence was clinically occult in 98 (60%) and clinically evident in 64 (40%). Clinically evident recurrences were more often superficial (skin, subcutaneous, or nodal) or in the brain, whereas clinically occult recurrences more often visceral. Overall, 90% of all recurrences were detected by 2.8 years. In all, 70% of patients with recurrence underwent curative-intent salvage treatment (locoregional: 94%, distant: 48%), with similar rates for clinically occult versus clinically evident recurrences (66 vs. 75%, P=0.240). Overall survival was superior among those who underwent curative-intent salvage treatment [5.9 vs. 1.2 years; hazard ratio=4.27, 95% confidence interval (CI)=2.68-6.80; P<0.001], despite 79% developing recurrence again. PET/CT had high sensitivity (88%, 95% CI=79.94-93.31%), specificity (90%, 95% CI=88.56-91.56%), and negative predictive value (99%, 95% CI=98.46-99.52%). However, the positive predictive value was only 37% (95% CI=31.32-43.68%). In patients with resected stage III-IV melanoma, surveillance PET/CT detected a large proportion of clinically occult recurrences amenable to curative-intent salvage treatment. Despite a high rate of second relapse, curative-intent salvage treatment was associated with superior overall survival. Even though PET/CT had high sensitivity, specificity, and negative predictive value, positive predictive value was poor, highlighting the need for histologic confirmation of PET/CT-detected abnormalities.