Although habitat fragmentation is a major threat to global biodiversity, the demographic mechanisms underlying species loss from tropical forest remnants remain largely unexplored. In particular, no studies at the landscape scale have quantified fragmentation's impacts on colonization, extinction, and local population growth simultaneously. In central Amazonia, we conducted a multiyear demographic census of 292 populations of two leaf-inhabiting (i.e., epiphyllous) bryophyte species transplanted from continuous forest into a network of 10 study sites ranging from 1, 10, and 100 to 110,000 ha in size. All populations experienced significantly positive local growth ( ) and a nearly constant per-generational l 1 1 extinction probability (15%). However, experimental leaf patches in reserves of ≥100 ha experienced nearly double (48%) the colonization probability observed in small reserves (27%), suggesting that the proximate cause of epiphyll species loss in small fragments (≤10 ha) is reduced colonization. Nonetheless, populations of small fragments exhibit rates of colonization above patch extinction, positive local growth, and low temporal variation, which are features that should theoretically reduce the probability of extinction. This result suggests that for habitat-tracking metapopulations subject to frequent and stochastic turnover events, including epiphylls, colonization/extinction ratios must be maintained well above unity to ensure metapopulation persistence.