Loss of native vegetation cover is thought to be a major driver of declines in pollination success worldwide. However, it is not well known whether reducing the fragmentation of remaining vegetation can ameliorate these negative effects. We tested the independent effects of composition vs. configuration on the reproductive success of a keystone tropical forest herb (Heliconia tortuosa). To do this we designed a large-scale mensurative experiment that independently varied connected forest-patch size (configuration) and surrounding amount of forest (composition). In each patch, we tested whether pollen tubes, fruit, and seed set were associated with these landscape variables. We also captured hummingbirds as an indication of pollinator availability in a subset of patches according to the same design. We found evidence for an effect of configuration on seed set of H. tortuosa, but not on other aspects of plant reproduction; proportion of seeds produced increased 40% across the gradient in patch size we observed (0.64 to > 1300 ha), independent of the amount of forest in the surrounding landscape at both local and landscape scales. We also found that the availability of pollinators was dependent upon forest configuration; hummingbird capture rates increased three and one-half times across the patch size gradient, independent of forest amount. Finally, pollinator availability was strongly positively correlated with seed set. We hypothesize that the effects of configuration on plant fitness that we observed are due to reduced pollen quality resulting from altered hummingbird availability and/or movement behavior. Our results suggest that prioritizing larger patches of tropical forest may be particularly important for conservation of this species.