With the disadvantage of computed tomography showing cumulative biological effects and magnetic resonance imaging posing a problem in clinical availability and cost, several authors described the technique of ultrasonography to measure non-invasively local cross-sectional dimensions (LCSDs) of masseter muscle sites. However only few studies addressed the issue of 'technique-related factors for intra- and inter-observer reliability' to gain more consistent testing and diagnosis. The purpose of the present study was to determine (1) whether the scanning level and/or the muscle condition may affect LCSD measurements and (2) whether measurements made at identical levels may be reproducible. The study included 35 subjects with signs and symptoms of temporomandibular disorders. Bilateral ultrasonographic investigation was performed with a linear (B-scan) 7.5 Mhz small-part transducer to register LCSDs of the anterior masseter muscle on three different levels. Scans were made on the relaxed and contracted muscle. Measurements were made in two sessions with a time interval of at least 5 min. Data were analysed for reproducibility by using the intra-class correlation coefficient (ICC) and the method error (ME). Scanning level and muscle condition had a significant effect on muscle measurements (P = 0.000). There was no difference in LCSD between the right and left muscle (P = 0.531). Measurements recorded at a given site were consistent across the testing sessions (P = 0.058). The scanning level with highest reproducibility was halfway between the origin and insertion (ICC = 0.92; ME = 6.2%). The data suggest that ultrasonography is a reliable method for measuring LSCDs of the anterior masseter muscle.