Studies of plants in fragmented habitats have focused on single ecological processes, such as seed Production or seed dispersal, that may be altered by increased fragmentation and that operate as the mechanism(s) that increase extinction probability. We examined a suite of potential mechanisms to explain demographic shifts toward extinction in populations of Trillium ovatum, a long-lived herbaceous pcr perennial found in the understory of western North American conifer forests. Past work has shown that populations of T. ovatum within similar to 65 nz of forest-clearcut edges in southwestern Oregon have had almost no new recruitment since the edges were formed We hypothesized that changes in abiotic conditions and biotic interactions present along edges are responsible for reduced recruitment. In eight populations in eight separate fragments, we evaluated the relationship of distance of the populations to the forest edge with respect to six processes: flowering phenology (timing), seed production, pollination- and resource-limitation of seed set, seed dispersal, seed predation, and germination. Those factors that showed a significant relationship with edge distance were then compared with recruitment of younger age classes. Two processes were significantly different near edges and were highly correlated with decreased recruitment: decreased seed production due to changes in pollination and increased seed predation by rodents. Our study (in conjunction with previous studies) suggests that several ecological processes show no significant relationship with edge distance and can be eliminated as possible mechanisms of reduced recruitment: flowering phenology, resource-limitation of seed set, seed dispersal, germination, herbivory, and survivorship of established plants. Edges influence some but not all components of a plant's life history. Thus, determining shifts in only one part of a life history will be inadequate for testing the prediction of increasing extinction probabilities in fragmented landscapes. Future studies should include enough information to conduct comprehensive analyses, such as matrix projections and sensitivity analyses.