The purpose of this study was to compare reproducibility in the assessment of caries lesion behaviour in digital subtraction images and conventional radiographs. Ninety–seven pairs of conventional film bite–wings were included. The radiographs had been taken with a Kwik–Bite filmholder without further standardisation. The selection criterion was that at least one surface imaged in a bite–wing pair should show progresssion of a caries lesion from the first exposure to the second. The time interval between the bite–wings was 1–2 years. The radiographs were scanned into a personal computer and the two images from the same patient subtracted. Seven observers scored the subtraction images and the conventional films. The scale was based on no change/ change in surface appearance scored in the following categories: 0 = no change; 1 = development of a new caries lesion in enamel; 2 = development of caries lesion in enamel and dentine/progression of an existing enamel lesion into dentine/progression of a dentinal lesion further into dentine; 3 = development of secondary caries; 4 = new filling/change of filling/extension of filling. Cohen’s κ coefficients were calculated for intra– and inter–observer agreement for the two radiographic methods separately. The κ value served as the statistical sampling unit for testing differences between the two methods (Wilcoxon’s test for ranked pairs). The average intra–observer κ value was 0.875 (range 0.775–0.958) for the subtraction images and 0.758 (range 0.560–0.890) for the conventional radiographs. The difference was statistically significant (p<0.05). For inter–observer agreement, the κ values were lower than for intra–observer agreement. The average value was 0.678 (range 0.485–0.754) for the subtraction images and 0.701 (range 0.497–0.817) for the conventional radiographs (p>0.1). This study suggests a new way of comparing reproducibility in terms of κ values between diagnostic methods in clinical studies where accuracy cannot be evaluated. The subtraction method may be useful in the assessment of caries lesion behaviour in the clinic.