Background: Most studies assessing the sensitivity and specificity of caries detection methods have been performed in vitro using the histological method as the gold standard showing inconsistent values. The aim of this study was to compare the sensitivity and specificity in detecting occlusal caries using the International Caries Detection and Assessment System (ICDAS II) with the radiographic method (RM), while using the Spectra™ Caries Detection System (SCDS) as the reference test. Methods: One hundred sixty children, ages 7–12 years, participated in the study. Five zones in the occlusal surfaces of 859 primary and 632 first permanent molars were examined visually using ICDAS-II, the RM using bitewing radiographs and SCDS. The descriptive statistics of sensitivity and specificity were calculated and compared. Results: For all molars combined and for primary molars only, the sensitivity of ICDAS II was higher for detecting total caries (p < 0.001), caries in enamel (p < 0.001), and caries in dentin (p = 0.016), but it was not different for detecting caries in the dentin of permanent first molars (p = 0.214), and primary second molars (p = 0.761). The specificity of RM was higher for detecting total caries, caries in enamel for all molars combined and for permanent first molars (p < 0.001). For caries in dentin, the specificity of ICDAS II was higher for all molars combined and for primary molars only (p < 0.001). For total caries in primary molars only, and caries in dentin of permanent first molars only, the specificity was not different (p = 0.156 and p = 0.181 respectively). Conclusions: The sensitivity and specificity of ICDAS II and RM changes depending on whether the carious lesion compromises the enamel or dentin, and if the caries detection is performed in the primary molars or permanent first molars.