In 1983 a classification scheme was proposed for patients with atypical naevi, according to their personal and family history of melanoma and atypical naevi. To assess the predictive value of these features we undertook prospective surveillance of patients at high risk of primary melanoma. We followed up 116 patients each with 3 or more clinically atypical naevi for at least 5 years. Patients are examined and naevi are photographed every 3-6 months; lesions showing disturbing change are excised for histopathology. Among 85 patients with no personal or family history of melanoma, 5 invasive (level 2 or deeper) melanomas developed during 583 person-years of follow up. The expected number of invasive melanomas in this population would be 0.054; the increased risk is significant (p < 0.001; relative risk 92 [95% CI 30-216]). There was a similarly increased risk of new melanoma also among 24 patients with atypical naevi plus a history of previous melanoma (observed 2, expected 0.022, p < 0.001; relative risk 91 [11-328]). By comparison, no second melanoma developed among 25 patients with previous melanoma but a normal naevus pattern during 213 person-years of similarly intensive follow-up. The risk of melanoma was highest among 7 patients with atypical naevi and a family history of melanoma (observed 6, expected 0.009, p < 0.001; relative risk 444 [121-1138]). The median thickness of surveillance-detected melanomas was 0.75 mm (range 0.40-1.05 mm) in this group. This study shows the value of clinical follow-up of high-risk patients to detect early thin melanomas.