Counts of animal carcasses are often used to estimate fatality caused by disease, environmental accidents (oil spills, radiation leaks), or human structures (power lines, sky scrapers, wind turbines). The need to adjust raw carcass counts for imperfect detectability to produce unbiased estimates of fatality has long been recognized, but the accuracy and precision of some estimators used to make the adjustments have not been evaluated. In this paper, I formalize a conceptual model of fatality and the factors that lead to imperfect detection, primarily removal by scavengers before searches can be carried out and inability of searchers to find all remaining carcasses. I propose an estimator of fatality that adjusts for imperfect detectability. Through simulation I evaluate the statistical properties (bias and precision) of this estimator and two others commonly used to estimate fatality at wind power facilities, when sources and magnitudes of imperfect detectability vary. None of the estimators was always unbiased under all conditions. Bias in the proposed estimator never exceeded ±27% whereas bias in the other two estimators was always negative and exceeded that of the proposed estimator in 98% and 93% of the simulated conditions, respectively. The proposed estimator was relatively robust to variation in sources and magnitudes of imperfect detectability, but was sensitive to distributional assumptions regarding carcass removal rates and searcher efficiency. It offers significant improvement over two current estimators and provides relatively unbiased estimates of fatality that can be applied under a variety of conditions and survey protocols.