Biologists have monitored black bear (Ursus americanus) populations using annual Lincoln-Petersen (L-P) estimates of population size derived from the fraction of marked bears recovered in the harvest. Although spatial, temporal, and demographic factors have been linked to variation in harvest rates of black bears, the effect of this heterogeneity on mark–recapture population estimates has not been evaluated. Failure to incorporate heterogeneity in harvest rates can result in biased population estimates and poor precision, which may lead to inappropriate management decisions. We used records of 6,982 bears captured during 1983–2001 in Pennsylvania, USA, to estimate the probability of harvest related to spatiotemporal, environmental, and demographic characteristics associated with individual bears. Harvest rates varied according to sex, age class, hunter density, and snow cover. In addition, harvest rates varied temporally (by year and month of capture) and spatially across Pennsylvania. Model selection based on Akaike's Information Criterion (AIC) supported a more complex harvest model based on a Horvitz-Thompson (H-T) estimator than the simpler model implicitly assumed by a series of annual L-P estimates. The H-T estimates of population size, which incorporated heterogeneity in harvest rates, were consistently lower than the annual L-P estimates because the H-T estimates accounted for mortality that occurred prior to the hunting season in addition to other sources of harvest heterogeneity. However, significant heterogeneity among breeding-age females could not be incorporated because we did not know the reproductive status (pregnant or with cubs) of each tagged and harvested female bear. Additional predictive variables of harvest rates of breeding-age females could further improve our model; however, an extension of our model that incorporates data from tagged cohorts may be an alternative means to improve the accuracy and precision of population estimates.