Plants and pollinators are experiencing parallel declines worldwide, despite theory predicting that pollination networks should withstand disturbance due to redundancy of pollinators, rarity of interactions between specialists, and flexible pollinator foraging behavior (re-wiring). Experiments evaluating the stability of plant-pollinator networks may help resolve this incongruity but remain uncommon. In this study, we simulated the extirpation of a hummingbird-pollinated understory plant, Heliconia tortuosa, from tropical forest fragments using a replicated Before-After-Control-Impact experimental design while quantifying hummingbird space use (383 hummingbird captures and 72 radio-tagged individuals), floral visitation rates (6,759 visitations from 20,725 observation hours), and plant pollination success (529 flowers). To complement this experimental approach, we also examined these responses across a natural gradient in H. tortuosa density. We expected that declines of H. tortuosa would either result in (a) network collapse, in which hummingbirds vacate fragments and compromise the reproductive success of other flowering plants, or (b) increased hummingbird reliance on alternative resources, leading to sustained fragment use. We also hypothesized that landscape and local context (i.e., connectivity to additional forest area and alternative resource abundance) could mediate hummingbird responses to H. tortuosa declines; for example, connectivity could facilitate hummingbird visitation to areas of reduced food availability. In our removal experiment, hummingbird persistence and plant pollination success were remarkably resistant to loss of H. tortuosa, a locally common plant species representing >40% of the available nectar resources on average. However, naturally low H. tortuosa densities were associated with reduced floral visitation rates and decreased pollination. Although landscape context (connectivity) led to higher hummingbird abundance, we found little evidence that connectivity or resource availability mediated hummingbird responses to declines in H. tortuosa availability. The exact mechanisms enabling short-term hummingbird persistence after resource removal remain unclear, as we did not discover evidence of rewiring. Physiological adaptations (e.g., torpor and insectivory), may have contributed to hummingbird resilience, perhaps alongside high movement abilities that facilitated visitation to spatially dispersed floral resources. With the important caution that short-term experiments may not realistically emulate natural extinction processes, our study provides partial support for theoretical predictions that pollination networks may be relatively robust to plant species loss.