Traditional approaches to the study of fragmented landscapes invoke an island-ocean model and assume that the nonhabitat matrix surrounding remnant patches is uniform. Patch isolation, a crucial parameter to the predictions of island biogeography and metapopulation theories, is measured by distance alone. To test whether the type of interpatch matrix can contribute significantly to patch isolation, I conducted a mark-recapture study on a butterfly community inhabiting meadows in a naturally patchy landscape. I used maximum likelihood to estimate the relative resistances of the two major matrix types (willow thicket and conifer forest) to butterfly movement between meadow patches. For four of the six butterfly taxa (subfamilies or tribes) studied, conifer was 3-12 times more resistant than willow. For the two remaining taxa (the most vagile and least vagile in the community), resistance estimates for willow and conifer were not significantly different, indicating that responses to matrix differ even among closely related species. These results suggest that the surrounding matrix can significantly influence the "effective isolation" of habitat patches, rendering them more or less isolated than simple distance or classic models would indicate. Modification of the matrix may provide opportunities for reducing patch isolation and thus the extinction risk of populations in fragmented landscapes.