Many ecologists believe birds disappear from tropical forest fragments because they are poor dispersers. We test this idea using a spatially explicit capture data base from the Biological Dynamics of Forest Fragments Project near Manaus, Brazil. We measure bird movements directly, over relatively large scales of space and time, both before and after landscape fragmentation. We found that species which disappear from fragments move extensively between plots before isolation, but not after, and often disperse to longer distances in continuous forest than in fragmented forest. Such species also preferentially emigrate from smaller to larger fragments, showing no preference in continuous forest. In contrast, species that persist in fragments are generally less mobile, do not cross gaps as often, yet disperse further after fragmentation than before. ‘Heavy tailed’ probability models usually explain dispersal kernels better than exponential or Gaussian models, suggesting tropical forest birds may be better dispersers than assumed with some individuals moving very long distances.