Despite the questionable sensitivity of conventional radiographic evaluation for slowly progressing proximal caries, application of digital subtraction imaging, with its much-improved performance for revealing subtle changes in mineralization, has not seen general clinical acceptance. This reluctance may be attributed to the non-trivial user-involvement required in the retrospective correction for discrepancies in projective geometry between serial radiographs. The aim of this study was to compare the performance of a mutual information based automated method for retrospective geometric standardization, with a manual reference point method, and with a automatic extraction of reference points method.