Sex determination of the human pelvis has traditionally been done through visual analyses of morphoscopic traits and there are limited metric methods available to forensic anthropologists to add metric credibility to these analyses. The goal of this research was to create an improved metric method using three-dimensional geometric morphometrics to determine sex from both whole and modeled fragmented human pubic bones. The sample consisted of n = 378 pubic bones from the University of New Mexico’s Maxwell Museum Documented Skeletal Collection and eight landmarks were collected from each bone. Statistical analyses and machine learning algorithms were used to predict the accuracy of the method’s ability to classify a bone as male or female on both whole and simulated fragmented remains; this included tests run on each possible landmark combination of three or more landmarks to simulate fragmented bones (218 combinations). The results of the whole bone analysis resulted in 95.35% testing accuracy. The results of the modeled fragmentary analysis consisted of 164 combinations which exhibit a 90% or higher accuracy in sex prediction; and twelve combinations which exhibit 96% or higher accuracy in sex prediction. In particular, two landmarks clustered around the ventral arc of the pubic bone performed the best, indicating this is the most sexually dimorphic portion of the bone. These results indicate that three-dimensional geometric morphometrics is a valid method to be applied to sex determination in forensic anthropology.