Although the number of genetic markers available for fisheries research has steadily increased in recent years, there is limited information on their relative utility. In this study, we compared the performance of different "classes" of genetic markers (mitochondrial DNA (mtDNA), nuclear DNA (nDNA), and allozymes) in terms of estimating levels and partitioning of genetic variation and of the relative accuracy and precision in estimating population allocations to mixed-stock fisheries. Individuals from eight populations of fall-run chum salmon (Oncorhynchus keta) from the Yukon River in Alaska and Canada were assayed at 25 loci. Significant differences in mitochondrial haplotype and nuclear allele frequencies were observed among five drainages. Populations from the U.S.-Canada border region were not clearly distinguishable based on multilocus allele frequencies. Although estimates of total genetic diversities were higher for the DNA loci (H_t = 0.592 and h = 0.647 for nDNA and mtDNA, respectively) compared with protein allozymes (H_t = 0.250), estimates of the extent of population differentiation were highly concordant across marker classes (mean theta = 0.010, 0.011, and 0.016 for allozymes, nDNA, and mtDNA, respectively). Simulations of mixed-stock fisheries composed of varying contributions of U.S. and Canadian populations revealed a consistent bias for overallocation of Canadian stocks when expected Canadian contributions varied from 0 to 40%, due primarily to misallocations among genetically similar border populations. No single marker class is superior for differentiating populations of this species at the spatial scale examined.