The purpose of this study was to determine the validity of detecting approximal imperfections of composite fillings using three intraoral radiographic systems in vitro. Class II composite resin restorations (108) with three radiopacities (264, 306, 443% Al 99.5) of which 27 had marginal openings or overhangs, respectively, were conventionally (Ektaspeed plus) and digitally (Dexis, Digora) radiographed. Images were assessed by 10 observers for the presence of marginal gaps and overhangs, as well as for their need of restorative treatment according to a five-point confidence rating scale. The validity of the observations were expressed as areas under receiver operating characteristic (ROC) curves (Aroc). Repeated measures analysis of variance revealed significant effects of 'radiographic system' and 'diagnostic purpose'. Marginal overhangs (Aroc = 0.90) were significantly easier to diagnose than openings (Aroc = 0.63). Marginal gaps were better detected on conventional and Dexis radiographs than on Digora images. the range of sensitivities and specificities of the treatment decision was 0.53-0.56 and 0.87-0.88, respectively. It was concluded that the validity of detecting marginal defects of composite resin restorations based on radiographs was only slightly affected by the radiographic system being used. The diagnosis of marginal gaps frequently resulted in false-positive and false-negative decisions.