Adjusting existing methods of human identification developed by forensic anthropologists in the United States for use with populations not included in the original development of an analytical method requires data collection using contemporary osteological collections from those populations, and an assessment of the within-group variation present. The primary purpose of this research is to document cranial macromorphoscopic trait variation using methods previously developed in the United States in a sample of 244 individuals from Antioquia, Medellín, Colombia. All individuals are of known age, sex, and birth region. The complex population and demographic history of Colombia makes ancestry assessment particularly difficult in that country. To that end, we explore inter-regional variation throughout Antioquia using birthplace to determine whether forensic anthropologists can provide finer levels of detail beyond identifying an unknown set of human remains as 'Colombian' or, more generally, Hispanic. State and local levels of identification resulting from the varied population histories of each state within Antioquia permit finer resolution, but only to a degree of certainty. Artificial neural networks (aNN) correctly classified only 18.6% of a validation sample, following modest classification accuracies of test/tuning (11.6%) and training (82.8%) samples to original birthplace. As with most neural networks, overfitting is an issue with these analyses. To remedy this overfitting and to document the applicability of aNNs to the assessment of ancestry in Colombia, we pooled the sample of Colombian data and compared that to modern American samples. In those analyses, the best aNN model correctly classified 48.4% (validation) of the sample. Given these results, finer levels of analysis in Colombia are not yet possible using only macromorphoscopic trait data.