The aim of this study was to compare the accuracy of limited cone beam computed tomography (LCBCT), an image plate system and F-speed film in assessing the depth of proximal carious lesions. Radiographs of a dry mandible with sound and carious teeth were obtained with all three methods. In 41 molar and premolar proximal surfaces, 2 observers independently measured lesion depth on the images from the three modalities. The correlation of measurements was assessed with Pearson's correlation analysis. Results from imaging modalities and histological sectioning (gold standard) were compared using Bland-Altman plots. Overall comparison of the depth measurements from the imaging modalities and the gold standard was done using repeated-measures ANOVA. Pairwise comparisons of systems were done by the Bonferroni t test. The correlation between the measurements of the two observers was 0.977 for film, 0.997 for image plate system and 0.998 for LCBCT. Bland-Altman plots revealed that LCBCT agreed very closely with the gold standard while the agreement between the latter and the image plate - or film - images was moderate. The mean difference and 95% limits of agreement between LCBCT and the gold standard were smaller than those between either image plate or F-speed film and the gold standard. The LCBCT method appears as a promising tool for detection and monitoring of proximal carious lesions.