Abundance estimators that account for imperfect detection, such as N-mixture models, assume that detection of individuals is independent of abundance. Using spot-mapping and N-mixture models applied to point-count data, we estimated abundance of Goldencheeked Warblers (Setophaga chrysoparia) in two years at six study sites at the Balcones Canyonlands Preserve, Austin, Texas. N-mixture model estimates deviated from spot-mapping estimates at the site level by overestimating at low abundances, and at the survey-station level by underestimating at high abundance, which suggests that model assumptions may have been violated. We tested whether detection of individuals is influenced by abundance by assessing per capita song rate in relation to abundance. Per capita song rate increased with abundance, illustrating how the behavior of a territorial passerine may violate the independent-detectability assumption. We next explored violation of this assumption at the survey-station level by applying N-mixture models to simulated data exhibiting heterogeneity in detection. This exercise revealed a slight but increasingly negative bias (underestimation of abundance) in the estimator as the actual abundance increased, given positive density-dependent detection. The simulations also revealed a potentiel effect of sampling variation on misestimation by N-mixture model estimators. Assessing the strength, basis, and prevalence of density-dependent detection; further analyzing the effects of nonrandom heterogeneity in producing estimator bias; and accounting for nonrandom detection heterogeneity in abundance estimators are fruitful areas for further study.