Accurate analysis of trends in densities estimated from survey data depends on selection of a model that fits the data. When such analysis uses data recorded at different times in the same area but from different transects, then the target species' density and spatial heterogeneity become an issue. It is possible, even in the absence of environmental change, that one set of transects may underestimate or overestimate the density entirely on the basis of location. The resulting risk is that the density estimated earlier may falsely indicate a decline or an increase when compared to the densities estimated later. We investigate this problem for trend analysis of two species of endangered Hawaiian birds at Hakalau Forest National Wildlife Refuge, island of Hawaii. Assertions based on Bayesian regression, and accepted by the U.S. Fish and Wildlife Service, purportedly indicate increases in the density of the endangered Hawaii Akepa (Loxops coccineus coccineus) from 1977 to 2007 and of the endangered Hawaii Creeper (Oreomystis mana) from 1987 to 2007. Here we show, at four study sites spanning the south—north axis of the refuge, that capture rates per mist-net hour of both species declined significantly and the 1977 transects used underestimated density because of location. We submit that the main reason for the differences between our piecewise regression and Bayesian regression is the inappropriate use of Bayesian regression. Analysis by the appropriate model indicates that since 2000 all Hawaiian birds at Hakalau Forest National Wildlife Refuge have been declining.