We tested forecasts made by Mac Nally and Bennett (Mac Nally, R., Bennett, A.F., 1997. Species-specific predictions of the impact of habitat fragmentation: local extinction of birds in the box-ironbark forests of central Victoria, Australia. Biological Conservation 82, 147–155) of relative vulnerability to habitat fragmentation for 43 species of birds in the box–ironbark system of central Victoria, Australia. The predictions were based on a simple, tripartite model linking habitat specialization, density and mobility, three of the characteristics most widely mooted in the literature to influence vulnerability to habitat fragmentation. For each species, a predicted index of ‘proneness’ to local extinction was calculated by using prior, independent data. The model system consisted of existing fragments of 10, 20, 40 and 80 ha size-classes, with between 5 and 15 ‘replicates’ of each size-class. Replicated ‘reference-areas’ — mapped-out areas of these same size-classes set within large (>10 000 ha) remnant forests of the same general vegetation as the fragments — were surveyed also to provide ‘expectations’ of the likelihood of finding each species in a given size-class. Our results indicated that the simple model had virtually no predictive power: the index-values for proneness to local extinction were no help in forecasting whether a given species was over- or under-represented in a fragment of a given size based on the bird's occurrence in reference-areas of similar size. When the three components of the model were tested separately, measures of habitat specialization and mobility showed no relationship with the observed responses of birds to fragmentation, while for population density there was a significant relationship but in a direction contrary to that expected. Evaluation of assumptions underlying the test revealed two main factors that are likely to have affected the model's performance: (1) differences in anthropogenic impacts between fragments and reference-areas have resulted in differences in habitat quality, and (2) species interactions within fragments differ from those in reference-areas, a result of secondary effects arising from the altered spatial pattern of the habitat (e.g. the expansion of numbers and influence of ‘despotic’ species such as the noisy miner Manorina melanocephala). These two factors appear to be common to most fragmented systems and, therefore, need to be addressed in predictive models of species vulnerability. However, because the nature of anthropogenic impacts and the types of changes to species interactions in fragments differ between ecosystems, we conclude that it will be difficult to develop a model of species' vulnerability to fragmentation that has both strong predictive ability and wide generality among ecosystems.