We investigated the role of matrix type as a determinant of change in bird densities with forest patch area (patch area effect) in two different Fennoscandian landscape types: mature forest fragments surrounded by cut-over or regenerating forest and true forested islands surrounded by water. Since the matrix of forested archipelagoes offers no resources to and impedes movement of forest birds, we predict that patch area effects on bird densities should be stronger on forested islands than in forest patches fragmented by forestry. We compiled correlation estimates of the bird density-patch area relationship from the literature and analyzed the data using meta-analysis. Combined correlation coefficients were significantly positive on islands but were not significantly different from 0 in fragments. Within-species comparisons also showed that correlations were consistently more positive on islands than in fragments. On islands but not in fragments, the densities of forest specialist species were more sensitive to area than were the densities of forest generalists, suggesting that specialists are more sensitive to changes in matrix quality. Migration status was only weakly associated with bird responses to island or fragment area. Thus, forest fragments do not function as true islands. We interpret this as the result of compensatory effects of the surrounding matrix in terms of availability of resources and enhanced connectivity (matrix quality hypothesis). A purely patch-centered approach seems an unrealistic framework to analyze population processes occurring in complex landscapes. The characteristics of the habitat matrix should therefore be explicitly incorporated into the assessment of species' responses to habitat fragmentation.