The forensic genetics field is generating extensive population data on polymorphism of short tandem repeats (STR) markersin globally distributed samples. In this study we explored and quantified the informative power of these datasets to addressissues related to human evolution and diversity, by using two online resources: an allele frequency dataset representing 141populations summing up to almost 26 thousand individuals; a genotype dataset consisting of 42 populations and morethan 11 thousand individuals. We show that the genetic relationships between populations based on forensic STRs are bestexplained by geography, as observed when analysing other worldwide datasets generated specifically to study humandiversity. However, the global level of genetic differentiation between populations (as measured by a fixation index) isabout half the value estimated with those other datasets, which contain a much higher number of markers but much lessindividuals. We suggest that the main factor explaining this difference is an ascertainment bias in forensics data resultingfrom the choice of markers for individual identification. We show that this choice results in average low variance ofheterozygosity across world regions, and hence in low differentiation among populations. Thus, the forensic geneticmarkers currently produced for the purpose of individual assignment and identification allow the detection of the patternsof neutral genetic structure that characterize the human population but they do underestimate the levels of this geneticstructure compared to the datasets of STRs (or other kinds of markers) generated specifically to study the diversity of humanpopulations.