Purpose: The directional dependence of accuracy with cone beam computed tomography (CBCT) has not been investigated thoroughly. The purpose of the present study was to clarify the effects of measurement direction and of phantom locations in the fields of view (FOVs) on the accuracy of linear measurement and on the limits of measuring thin objects with CBCT. Materials and Methods: An aluminum phantom was scanned by CBCT. The thickness in both the longitudinal and horizontal directions (LD and HD) was measured at both the center and periphery in the three FOVs. The length was determined by a 50% relative threshold method to eliminate observer-dependent measurement errors. The measurement limits of a thin object were assessed in thin parts of the phantom (thickness, 0.3 to 1.0 mm). Results: The measurement accuracy in the LD was excellent (within 1 pixel), while that in the HD was fair (0 to 2.35 pixels difference); a slight overestimation was found near the center of the phantom (1.06 to 2.35 pixels difference) in comparison to that of the edges (within 1 pixel difference). The overestimation was more marked at the longitudinal periphery than at the center of each FOV. A thickness of at least three to four pixels was necessary to keep errors within the one-pixel range along both directions. Conclusions: The accuracy of linear measurement with CBCT is excellent, especially when measuring in the LD. Distances are slightly overestimated in the HD near the center of the phantom. Close attention is therefore necessary when planning the placement of implants adjacent to horizontal vital structures. Second, a thickness of at least 3 to 4 pixels was necessary to maintain high accuracy for linear measurement. This must be taken into consideration when measuring a minute structure, such as thin cortical bone. Keywords: cone angle, cone beam computed tomography, dental implants, directional dependence, measurement accuracy, thresholds