Purpose: To compare linear measurements made on cone beam computed tomography (CBCT) images to direct measurements in ex vivo porcine mandibles. Materials and Methods: Six cross-sectional planes were defined by gutta-percha-filled cavities in seven mandibles. The mandibles were scanned in a CBCT apparatus and later sectioned, using a band saw, through the gutta-percha markers. Next, four direct (DIR) linear measurements were performed for each section with a handheld digital caliper, using the guttapercha markers as reference points. The corresponding radiographic (RAD) cross sections were then measured using dedicated software. A total of 168 sites were measured. Differences between RAD and DIR measurements [Δ(RAD - DIR)] were calculated for each pair individually. Results: Mean Δ(RAD - DIR) was -0.17 ± 0.53 mm (range, -1.42 to 1.09 mm). CBCT overestimated direct measurements at 36% of the sites; 8% of sites (95% confidence interval, 3.8% to 12.2%) showed errors between +0.5 and +1 mm, and 1.8% (95% confidence interval, -0.2% to 3.9%) showed errors greater than +1 mm. Conclusions: Good correlation was found between CBCT and direct measurements. However, the significant percentage of sites with overestimation of at least 0.5 mm indicates a need for safety margins to be maintained when CBCT is used to plan surgical interventions such as dental implant therapy. Keywords: accuracy, cone beam computed tomography, dental implants, dimensional measurement, ex vivo