OBJECTIVE We evaluated the influence of surrounding anatomical structures on the registration accuracy of a diagnostically important region, by varying the ROI (region of interest) window used in an automatic ROI-based digital subtraction method. METHODS Pairs of radiographs were taken at 2 molar regions using paralleling techniques of XCP devices (Rinn Co., Elgin, Ill) attached with and without a bite block. The global and local registration accuracies were measured by the RMS (root mean square) of subtraction images for various ROIs. RESULTS No significant differences were found in global registration accuracies among various ROI sizes in all the paralleling techniques. The local registration accuracies decreased with extension of the ROI size. CONCLUSIONS Registration using an ROI restricted to the anatomical region of diagnostic interest provides higher accuracy than using a larger ROI.