Estimating species richness (i.e., the actual number of species present in a given area) is a basic objective of many field studies carried out in community ecology and is also of crucial concern when dealing with the conservation and management of biodiversity. In most studies, the total number of species recorded in an area at a given time is taken as a measure of species richness. Here we use a capture–recapture approach to species richness estimation with North American Breeding Bird Survey (BBS) data in order to estimate species detectability and thus gain insight about its importance. In particular, competing models making different assumptions about species detectability are available. We carried out analyses on all survey routes of four states, Arizona, Maryland, North Dakota, and Wisconsin, in two years, 1970 and 1990. These states were chosen to provide contrasting habitats, bird species composition, and survey quality. We investigated the effect of state, year, and observer ability on the proportions of different models selected, and on estimates of detectability and species richness. Our results indicate that model M_h, which assumes heterogeneous detection probability among species, is frequently appropriate for estimating species richness from BBS data. Species detectability varied among states and was higher for the more skilled observers. These results emphasize the need to take into account potential heterogeneities in detectability among species in studies of factors affecting species richness.