1. Increasing numbers of studies are analysing the shapes of objects using geometric morphometrics with tomographic data, which are often segmented and transformed to three-dimensional (3D) surface models before measurement. The present study aimed to evaluate the effects of different image segmentation methods on geometric morphometric data collection using computed tomography data collected from non-human primate skulls. 2. Three segmentation methods based on a visually-selected threshold, a half-maximum height protocol and a gradient and watershed algorithm were compared. For each method, the efficiency of surface reconstruction, the accuracy of landmark placement and the level of variation in shape and size compared with various levels of biological variation were evaluated. 3. The visual-based method inflated the surface in high-density anatomical regions, whereas the half-maximum height protocol resulted in large numbers of artificial holes and erosion. However, the gradient-based method overcame these issues and generated the most efficient surface model. The segmentation method used had a much smaller effect on shape and size variation than interspecific and inter-individual differences. However, this effect was statistically significant and not negligible when compared with intra-individual (fluctuating asymmetric) variation. 4. Although the gradient-based method is not widely used in geometric morphometric analyses, it may be one of the most appropriate options for reconstructing 3D surfaces. When evaluating small variations, such as fluctuating asymmetry, care should be taken around combining 3D data that were obtained using different segmentation methods.