Radiographs of occlusal (n = 20) and approximal (n = 24) surfaces of extracted teeth were examined by 240 dentists before participating in continuing education courses dealing with caries diagnosis and treatment decisions. The radiographic caries diagnoses were treated in accordance with the receiver operating characteristic (ROC) technique, in which the area beneath the ROC curve (A, value) indicates the quality of the observations. The frequencies of false positives made in dentin radiographically were higher for approximal (20.7%) than for occlusal caries (12.3%). The quality of pooled radiographic diagnoses of occlusal dentin lesions for all observers was significantly better than diagnoses of approximal caries in dentin. A statistically significant relationship between the observer's qualities of diagnosis of caries on approximal and occlusal surfaces (P = 0.045) was found. For diagnosis of dentin caries on approximal surfaces the mean Cohen kappa was 0.74 (standard deviation (s), 0.12; range, 0.39-0.95), and the corresponding values for occlusal surfaces were 0.70 (s, 0.14; range, 0.25-0.98). In the material under study the dentists were at least as good at diagnosing dentin caries occlusally as approximally. To avoid overtreatment, the observer's diagnostic threshold should ideally be adjusted towards strict criteria when a positive diagnosis is synonymous with a filling. The diagnostic thresholds were stricter in diagnosing occlusal surfaces than for approximal surfaces, indicating a more optimal strategy among dentists while diagnosing occlusal dentin lesions in a population with low caries prevalence.