An increasing number of studies determine endocranial capacity (ECC) digitally using serial computed tomographic (CT) scan data. However, the multiple causes of errors inherent in such measurements and the resulting degree of accuracy have not yet been fully examined. In the present study, five observers estimated the ECCs of two modern human crania by segmentation of CT data (voxel size 0.380 mm) and by means of the conventional millet seed method. The ECC estimations were much more consistent in the CT-based (technical error, TEM = 1.4 cc) than the millet seed (TEM = 11.8 cc) methods. The estimated capacities also tended to be larger in the latter, suggesting possible systematic bias. Next, the causes and degree of error in the CT-based method were examined. Error due to size calibration of the CT images was no more than 3 cc (± 1.5 cc) per 1000 cc. Errors involved in delineating the endocranial cavity were evaluated as follows. First, we adopted a segmentation routine in which a single global threshold value was applied to most of the volume, and then supplemented locally by more appropriate values in the regions where the global value was insufficient. We then estimated the potential error introduced by choosing a global threshold value. We found this to be within 5 cc (per 1000 cc). This error range increased to 7.5 cc (per 1000 cc), when using a lower resolution data set (slice thickness 1.14 mm). Another source of interobserver error involves the blocking of foramina and canals, which we estimated to be less than 2 cc (per 1000 cc). Adding these various effects, our investigations indicate that ECC estimations based on high-resolution CT data are accurate to within ± 5 cc per 1000 cc. Finally, we reevaluated the ECC of the Minatogawa 1 skull to be 1335 cc (with a probable range of 1327–1343 ± 5 cc), considerably smaller than the originally reported value of 1390 cc obtained by the millet seed method.