The effect of population subdivision on estimated match probabilities has been raised [Nature 339 (1989) 501; Am. J. Hum. Genet. 48 (1991) 819; Science 254 (1991) 1921]. Previous work [J. Forensic Sci. 39 (1994) 319; J. Forensic Sci. 39 (1994) 988; Am. J. Hum. Genet. 55 (1994) 533] has compared product rule estimates from differing databases and found that the "subpopulation" error may be of the order of a factor of 10. This approach compares an estimate with an estimate. This paper uses simulation to extend these studies by allowing a comparison to a 'true match probability' and supports the conclusion that subpopulation effects are mild. In addition the performance of recommendations 4.1 and 4.2 of NRC II [National Research Council and C.O.D.F. Science, The Evaluation of Forensic DNA Evidence, National Academy Press, Washington, DC, 1996].