Intra- and interexaminer reproducibility of clinical caries diagnoses was studied using 3 experienced dentists, who independently examined 20 patients twice at an interval of 2–6 weeks. The lesions were classified as one of the following four diagnoses: (1) initial active caries, (2) initial inactive caries, (3) manifest active caries, and (4) manifest inactive caries. For the various diagnoses, signs of cavitation, surface structure and discolouration were used. The reproducibility at a patient level was evaluated using intraclass correlation coefficients and at a surface level using kappa statistics. At a patient level, for the total number of lesions, the intraclass coefficients of correlation varied from 0.55 to 0.77. For the individual diagnoses, total manifest caries showed the highest coefficients of correlation, ranging from 0.73 to 0.92. At a tooth surface level, the kappa values varied from 0.29 to 0.61. The most common shift was that from any type of caries diagnosis to a diagnosis of a sound surface. There was little difference between the results for coronal and root caries and between initial and manifest lesions. Nor was there any difference for buccal + lingual surfaces only compared to all surfaces. Judging from the pattern of change in diagnosis between the repeated examinations, the main source of error seemed to be due to the fact that the lesions observed at one examination were overlooked at the other examination.