Population estimates of white-tailed deer (Odocoileus virginianus) based on marking individuals and resighting them at later times require assumptions about observability that are rarely verified. Yet the accuracy of estimates depends on meeting the assumptions. At Remington Farms on the eastern shore of Maryland, we tested accuracy of mark-resight population estimates from the heterogeneity model (Mh) of the CAPTURE program against a known abundance of a marked subpopulation of the herd. We also tested the assumption of heterogeneous capture probabilities. We conducted evening road counts to resight animals marked with collars and used radio-telemetry observations to estimate sightability of individual animals. Estimates of observability were biased high, and concomitant population estimates were biased low by between 25% to 35%. Estimates determined from radio telemetry of the observabilities of different individuals ranged from 0.04 to 0.45, an 11-fold difference. The overall effect of a substantial heterogeneity in observabilities is probably a positively biased estimate of average observability and thus a negatively biased estimate of population abundance.