Using a TMJ skeletal phantom, we assessed validity and reliability of digitally subtracted linear tomographic images to quantify condylar position changes. Horizontally corrected frontal and lateral tomographic images were made with the condyle in a 'centred' position and displaced by 1, 2 and 3 mm increments in two of three directions (inferior, and posterior, or lateral). Film images were sequentially subtracted and then randomized so that observers were blinded to the amount and direction of condylar movement. Averages of three measurements of the dark bands representing condylar shifts were made using a digitizing tablet. The study utilized a nested design where the series of film images was subtracted twice and each subtraction was read twice by two examiners. We repeated the study with a second series of films treated in an identical manner. Average differences from expected values were similar for frontal and lateral techniques and all directions and increments of displacement. Typical average difference for any increment was less than 0.1 mm. Standard deviations in measures were similar across techniques and increments and were typically less than 0.1 mm. Significant variation in the absolute value of differences between measured and expected values by direction of condylar displacement was attributed to greater accuracy of registration on the vertical axis for frontal images. A significant source of variation in the study design element 'film set' was attributed to errors in repositioning the phantom.(ABSTRACT TRUNCATED AT 250 WORDS)