Background & Aim: The majority of carious lesions are not well-defined radiolucencies. Approximately 40% demineralization is required for radiographic detection of a lesion. The actual depth of penetration of carious lesion is deeper than may be detected radiographically. However, digital subtraction images permit to detect 1-5% decrease of mineral mass per unit volume. The goal of the present study was to evaluate the accuracy of digital subtraction radiography in the detection of dental demineralization in vitro.Methods & Materials: This study was based on observational-diagnostic method which was done on 30 extracted human teeth, categorized in two groups A and B, each having 15 members. In each of teeth, one approximal enamel demineralization lesion was induced using an acidified system (PH=4.8). Direct digital radiography were obtained under standardized condition of teeth before demineralization. After 7 days, the teeth of group B and after 42 days, group A removed from acid and new radiographs were taken. The images of the 7th and 42nd days were subtracted from the baseline radiograph (before creation of the lesion). Then teeth were histologically evaluated. Direct digital and subtraction images were interpreted by three observers to detect presence or absence of the lesion, then the diagnostic accuracy of both methods was determined.Results: After 7 days, the sensitivity, specificity, accuracy, positive predictive value and negative predictive value for incipient lesions in direct digital radiography were 0%, 80%, 40%, 0% and 44% respectively and in digital subtraction radiography were 66.7%, 86.7%, 76.7%, 83.4% and 72.3% respectively. However after 42 days the sensitivity, specificity and accuracy of both methods were 100%.Conclusion: Digital subtraction radiography has a fairly acceptable accuracy in detection ofthe incipient proximal lesions in comparison with DDR. For moderate proximal lesions DSR has the same accuracy as DDR.