OBJECTIVES To compare reliability for landmark identification on patient images from three-dimensional (3D) cone beam CT (CBCT) and digital two-dimensional (2D) lateral cephalograms. METHODS Ten lateral cephalometric digital radiographs and their corresponding CBCT images were randomly selected. 27 observers digitally identified 27 landmarks in both modes. The x- and y-coordinates for each landmark, indicating the horizontal and vertical positions, were analysed for interobserver reliability by comparing each measurement to the best estimate of the true value. Intraobserver reliability was also assessed. Linear models and intraclass correlation coefficients (ICCs) were used for analyses. RESULTS For interobserver reliability, the following locations were farther from the best estimate for 2D than 3D: x-location in subspinale (A-point), anterior tip of the nasal spine (ANS), L1 lingual gingival border and L1 root; y-location in porion, ramus point and orbitale; x- and y-locations in basion, condylion, midramus, sigmoid notch and U6 occlusal. 3D y-locations were farther in the gonion, L1 tip, sella and U1 tip. For intraobserver reliability, 2D locations were farther in y-locations in orbitale and sigmoid notch, and both x- and y-locations in basion. 3D locations were farther in the x-location in U1 labial gingival border and y-locations in L1 tip, L6 occlusal, menton and sella. For intraobserver ICCs, greater variations in 2D than 3D included: A-point, ANS, midramus, orbitale, ramus point, sigmoid notch and U1 root. CONCLUSIONS 3D imaging, as in CBCT, allows for overall improved interobserver and intraobserver reliability in certain landmarks in vivo when compared with two-dimensional images.