The principal method for estimating the abundance of bats in roosts is to count the number of bats exiting the roost at dusk (‘flyout counts’). We hypothesised that the accuracy and precision of flyout counts decrease non-linearly as the number of bats moving per unit of time increases, and that accuracy increases with observer experience. To test these hypotheses, we filmed grey-headed flying-foxes (Pteropus poliocephalus) exiting a roost in Melbourne on three consecutive evenings. The film was slowed and the number of flying-foxes flying-out in 30-s intervals was counted and assumed to be the true abundance. Thirteen other observers independently counted the number of flying-foxes flying-out in real time. We formulated our hypotheses into candidate models and compared support for these models using information-theoretic methods. Non-linear models had much greater support than linear models for all three flyouts. There was undercounting in two flyouts and overcounting in the third. There was good support for an effect of observer experience in one of the flyouts, but less support in the others. Precision declined as the true abundance increased in all three flyout counts. Our results indicate that accuracy, precision and observer effects vary with the dynamics of each flyout, and suggest that under some conditions flyout counts will often provide both inaccurate and imprecise estimates of abundance.