We examined some models to predict the species richness of forest birds by the landscape-level factors at urban woods in Osaka Prefecture, Japan. The environmental factors examined were area (A) and elongation (E) of woodlands, the distance to mountain (D), distance to the nearest woods (> 10ha) (D 10), and the proportion of woods (Pw) and field (Pf) within 25 km2 outside the parks. The species–area relationship at 28 parks was better fitted by the power function (r2=0.704) and by the logistic function (r2=0.696) than by the exponential function (r2=0.637). A woods that was planted 7 years ago had extremely few species, but there was no significant difference in species richness between woods < 50 years old and those older. We built Principal Component Regression (PCR) models to predict the species richness, because collinearity was detected between D and Pw. PCR, log[S/(43−S)] = −1.820 + 0.224logA − 0.0113 D + 0.133 logPw − 0.0588E (R2 = 0.939, n = 27) was estimated. The species–area relationship was caused by difference in the occurrence pattern of bird species. This was attributable to the differences in feeding habit or micro-habitat use. We estimated that birds living between the forest interior and the edge, such as Dendrocopos kizuki, required at least 20 ha of woodland.