The aim of this study was to assess landmark identification on digital images in comparison with those obtained from original radiographs. Ten cephalometric radiographs were selected randomly. Seven orthodontic residents identified 19 cephalometric landmarks on both the original radiographs and the digital images. To assess the concordance between landmarks identified on the original radiographs and on their digital counterparts, the x, y coordinates for each landmark in the 2 modalities were transformed with the identical references. The placement differences for 19 landmarks between 2 methods were calculated and their components in horizontal and vertical directions were analyzed respectively. Multivariate analysis of variance showed that the "cephalometric radiograph" and "landmark" variation had greater influence than that from "method" (landmark identification on digital / original radiograph). It was also noted that the differences of landmark identification between original radiographs and their digital counterparts were statistically significant. The landmarks with significant differences of horizontal component on the x-axis were Me, Gn, ANS, PNS, LIA. The differences were generally under 1 mm with the exception of Or, Me, PNS, LIA. The landmarks with significant differences of vertical component on the y-axis were Po, Or, Gn. The inter-observer error for each landmark in digital images was generally larger than that in the original radiographs. However, statistically significant differences of inter-observer errors between 2 modalities were only found for 4 of the 19 landmarks. These 4 landmarks, Po, Ar, ANS, and UM, should be scrutinized more carefully during potential applications of digital cephalometry.