Human harvests can select against phenotypes favoured by natural selection, and natural resource managers should evaluate possible artificial selection on wild populations. Because the required genetic data are extremely difficult to gather, however, managers typically rely on harvested animals to document temporal trends. It is usually unknown whether these data are unbiased. We explore our ability to detect a decline in horn size of bighorn sheep (Ovis canadensis) by comparing harvested males with all males in a population where evolutionary changes owing to trophy hunting were previously reported. Hunting records underestimated the temporal decline, partly because of an increasing proportion of rams that could not be harvested because their horns were smaller than the threshold set by hunting regulations. If harvests are selective, temporal trends measured from harvest records will underestimate the magnitude of changes in wild populations.