We determined breeding-site selection of Northern Goshawks (Accipiter gentilis) by comparing 33 breeding sites to 40 nonbreeding sites at the small (nest tree and vegetation structure within an 11.3-m radius of the nest) and the large scale (biomass of prey species and landscape structure within a 500-m radius from the nest) in the urbanized area of Kanagawa Prefecture, Japan. Goshawks selected primarily Japanese cedar (Cryptomeria japonica) trees over 50-cm diameter at breast height (DBH) as nest trees. Trees smaller than 35-cm DBH were avoided. We used univariate logistic regression to assess vegetation structure of the nesting area. At the small scale, canopy trees in the nest plots were larger and taller, with more canopy cover, than those in non-nesting plots, and nests were farther from human habitations. In addition, nest plots had fewer understory trees than non-nest plots. In multivariate logistic regression, the DBH of canopy trees, number of understory trees, and distance to human habitation were highest in importance. At the large scale, nest plots had a greater percentage of forest area, smaller percentage of built area, more forest edge facing open land, and more forest edge facing PCDRS (parks, cemeteries, developed land, recreational fields, and small-scale vegetated areas <0.01 km² in a matrix of ≥30% residential areas). In the multivariate model, the length of forest edge facing open land and the percentage of built area were of highest importance.