We investigated differences in the statistical power to assign parentage between an artificially propagated and wild salmon population. The propagated fish were derived from the wild population and are used to supplement its abundance. Levels of genetic variation were similar between the propagated and wild groups at 11 microsatellite loci, and exclusion probabilities were >0.999999 for both groups. The ability to unambiguously identify a pair of parents for each sampled progeny was much lower than expected, however. Simulations demonstrated that the proportion of cases in which the most likely pair of parents were the true parents was lower for propagated parents than for wild parents. There was a clear relationship between parentage assignment ability and the estimated effective number of grandparents of the progeny to be assigned. If a stringent threshold for parentage assignment was used, estimates of relative fitness were biased downward for the propagated fish. The bias appeared to be largely eliminated by either fractionally assigning progeny among parents in proportion to their likelihood of parentage or by assigning progeny to the most likely set of parents without using a statistical threshold.