Regression models were developed to predict relative bird abundance in a naturally heterogeneous landscape using patch and landscape spatial scales. Breeding birds were surveyed with point counts on 140 study sites in 1997 and 1998. Aerial photographs were digitized to obtain habitat patch information, such as area, shape, and edge contrast. Classified remote-sensing data were gathered to provide information on landscape composition and configuration within a 1-km2 area around the study sites. Stepwise multiple linear regression was used to develop 40 species-specific models within specific habitat types using patch and landscape characteristics. In 38 out of the 40 models, area of the habitat patch was first selected as the most important predictor of relative bird abundance. Variables related to the landscape were retained in 6 of the 40 models. In this naturally heterogeneous region, the landscape surrounding the patch contributed little to explaining relative bird abundance. The models were evaluated by examining how well they predicted relative bird abundance in a test set not included in the original analyses. The results of the test data were reasonable: >79% of the test observations were within the prediction intervals established by the training data.