Body mass estimation equations are generated from long bone cross-sectional diaphyseal and articular surface dimensions in 176 individuals and 12 species of hominoids and cercopithecoids. A series of comparisons is carried out to determine the best body mass predictors for each of several taxonomic/locomotor groupings. Articular breadths are better predictors than articular surface areas, while cross-sectional shaft strengths are better predictors than shaft external breadths. Percent standard errors of estimate (%SEEs) and percent prediction errors for most of the better predictors range between 10–20%. Confidence intervals of equations using sex/species means are fairly representative of those calculated using individual data, except for sex/species means equations with very low %SEEs (under about 10%), where confidence intervals (CIs) based on individuals are likely to be larger. Given individual variability, or biological “error,” this may represent a lower limit of precision in estimating individual body masses. In general, it is much more preferable to determine at least broad locomotor affinities, and thus appropriate modern reference groups, before applying body mass estimation equations. However, some structural dimensions are less sensitive to locomotor distinctions than others; for example, proximal tibial articular M-L breadth is apparently “locomotor blind” regarding body mass estimation within the present study sample. In other cases where locomotor affiliation is uncertain, mean estimates from different reference groups can be used, while for some dimensions no estimation should be attempted. The techniques are illustrated by estimating the body masses of four fossil anthropoid specimens of Proconsul nyanzae, Proconsul heseloni, Morotopithecus bishopi, and Theropithecus oswaldi. Am J Phys Anthropol 120:16–37, 2003. © 2003 Wiley-Liss, Inc.