Studies of detection and quantification of dental caries by KaVo DIAGNOdent have shown high diagnostic accuracy and high inter-operator agreement in vitro. The aim of the present study was to determine whether comparable results could be achieved under clinical conditions. In Part I (occlusal surfaces), 52 occlusal sites (suspected caries) were examined by visual inspection, bitewing radiography, and two DIAGNOdent devices. The fissures were then opened, and lesion depth established according to a four point scale. In Part II (smooth surfaces), two operators performed DIAGNOdent measurements of incipient carious lesions on the buccal surface of 30 molar teeth. The readings were compared with corresponding measurements by Quantitative Light-induced Fluorescence (deltaF). The results for occlusal surfaces disclosed very low correlation between lesion depth and readings by the two DIAGNOdent devices. The intra-operator agreements were very good (0.80-0.92), and the inter-operator agreement was good (0.71-0.87). A significant systematic difference in readings by the two devices precludes general recommendations of cut-off levels. However, individually suitable cut-off values for dentinal lesions gave good qualitative information about the surface (77%-85% over all correct observations), rather than quantitative information in terms of lesion depth within the dentine. In this context, the DIAGNOdent device might be of help for the clinician in the process of decision making about invasive therapy. For smooth surfaces, the correlation between readings obtained with the DIAGNOdent device and deltaF were acceptable (0.57-0.73). The intra-operator agreement was 0.94 for both operators (excellent), and the inter-operator agreement ranged from 0.79 to 0.87 (very good).