We used a bird-song simulation system to experimentally assess the effects of habitat, vegetation structure, and background noise on detection probability in aural avian point counts. We simulated bird songs of seven species in two habitats (mixed pine–hardwood forest and deciduous forest) and two leaf conditions (leaves on and leaves off) with two levels of background noise (~40 dB and ~50 dB). Estimated detection probabilities varied greatly among species, and complex interactions among all the factors existed. Background noise and the presence of leaves on trees decreased detection probabilities, and estimated detection probabilities were higher in mixed pine–hardwood forest than in deciduous forest. At 100 m, average estimated detection probabilities ranged from 0 to 1 and were lowest for the Black-and-white Warbler (Mniotilta varia) and highest for the Brown Thrasher (Toxostoma rufum). Simulations of expected counts, based on the best logistic model, indicated that observers detect between 3% (for the worst observer, least detectable species, with leaves on the trees and added background noise in the deciduous forest) and 99% (for the best observer, most detectable species, with no leaves on the trees and no added background noise in the mixed forest) of the total count. The large variation in expected counts illustrates the importance of estimating detection probabilities directly. The large differences in detection probabilities among species suggest that tailoring monitoring protocols to specific species of interest may produce better estimates than a single protocol applied to a wide range of species.