The problem of scaling is considered for community abundance distributions. Whereas species number scales predictably with spatial extent according to the species–area relation, it is shown that scaling of abundance distributions is not simple. To develop a scaling model, the structure of the bird community for the US Pacific and Inland Northwest was analyzed using Breeding Bird Survey (BBS) route data. For individual routes, it was found that the log-series model fit the observed distribution better than the lognormal. A tendency was observed for routes whose most common species was more abundant to have lower equitability (less even distribution of numbers of individuals across species), as has been observed in prior studies. Equitability increased with the number of habitats (vegetation types) along a route. At the regional scale, the community abundance distribution had an unusual bowed shape that does not correspond to any known distribution such as lognormal or log-series. With more than 2.5 million birds in the sample, the shape of this distribution cannot be attributed to sampling error. The unusual empirical regional distribution could be matched by a sampling model that assumed species abundances on individual routes to be samples drawn from a lognormal distribution, used the observed distribution of number of routes per species, and used an empirical trend for more widespread species to be more abundant on routes where they occur. The resulting sampling model fit the empirical regional distribution almost perfectly (R 2 =0.985). This provides a scaling model that relates route-level data to regional data.