OBJECTIVE To elucidate the accuracy in vitro of quantitative digital subtraction for direct digital dental radiography using a commercially available image analysis system. METHODS Images were corrected first for dark current and sensitivity variation among pixels of the CCD sensor (first correction), and corrected further for interimage variations due to the variations in exposure conditions (second correction). A copper stepwedge with six steps was used in the second correction. Three corrected images derived from three original images at each different exposure time were examined to confirm the reliability of the second correction. Correlation between bone mineral content and pixel value was assessed in subtracted images of an artificial bone phantom. RESULTS The first and second corrections significantly improved the intra- and inter-image variation respectively. The second corrected images obtained at three different exposure times showed almost the identical grey levels. An original image was transformed to a copper equivalent image with a high reliability by our original method. The pixel value in the subtracted image between two copper equivalent images of a bone phantom was linearly related to the bone mineral content. There was a slight overestimate at lower pixel values but it was accurate at higher pixel values. CONCLUSIONS Quantification of digital subtraction using this method is sufficiently reliable for its application to clinical practice to be evaluated.