The influence of differences in imaging geometry on the detection of small periodontal bone lesions was studied when using a digital subtraction technique. The results were compared with those from conventional radiographs. Reference radiographs were obtained on human mandibular jaws before small lesions were made in the marginal alveolar bone. Subsequent radiographs were taken with the same imaging projection as well as with changes of the angulation in either vertical, or horizontal direction, or both. 10 dentists served as observers, who compared the conventional radiographs taken before and after the lesions were made, and registered the presence or absence of lesions using a diagnostic rating scale. The conventional radiographs were digitized, and subtraction images made. These images were interpreted by the same observers. For each observer, image modality and projection, the resulting rates of true positive and false positive diagnoses were plotted on a graph. The area under the curve, obtained by connecting the paired values of true positive and false positive results associated with the different diagnostic ratings, was used as a measure of diagnostic accuracy. Statistically significant differences could be demonstrated between the results obtained from the subtraction images, whether produced from identical or non-identical pairs of conventional radiographs, and the results from the conventional radiographs obtained under ideal conditions.