The Connecticut Warbler (Oporornis agilis) is a rare and declining neotropical migrant that breeds in the north-central United States and south-central Canada. To better understand the species' habitat needs, we analyzed 371 observations of the Connecticut Warbler over 18 years at 86 sites in 28 stands of forest in northern Minnesota. We considered the habitat and landscape at three spatial scales (buffer radii of 100, 500, and 1000 m) and regressed combinations of habitat variables with two response variables, the Connecticut Warbler's abundance (the total number of individuals ever recorded at a site or stand, with a zero-inflated negative binomial distribution) and frequency (the number of years recorded out of 18, with logistic regression). From a subset of models retained on the basis of Akaike's information criterion, we calculated model-averaged predictions for each combination of buffer size and response variable. Models based on Connecticut Warbler frequency at the 1000-m buffer performed best in comparisons of model-averaged predictions to observed data. At the 1000-m scale, Connecticut Warblers were positively associated with a combination of large patches of upland coniferous and lowland black spruce forest and were negatively associated with upland deciduous forest. From these models, we mapped predicted breeding habitat for the Connecticut Warbler in the areas sampled in northern Minnesota.