A capacity to predict the effects of fire on biota is critical for conservation in fire-prone regions as it assists managers to anticipate the outcomes of different approaches to fire management. The task is complicated because species’ responses to fire can vary geographically. This poses challenges, both for conceptual understanding of post-fire succession and fire management. We examine two hypotheses for why species may display geographically varying responses to fire. 1) Species’ post-fire responses are driven by vegetation structure, but vegetation – fire relationships vary spatially (the ‘dynamic vegetation’ hypothesis). 2) Regional variation in ecological conditions leads species to select different post-fire ages as habitat (the ‘dynamic habitat’ hypothesis). Our case study uses data on lizards at 280 sites in a ∼ 100 000 km2 region of south-eastern Australia. We compared the predictive capacity of models based on 1) habitat associations, with models based on 2) fire history and vegetation type, and 3) fire history alone, for four species of lizards. Habitat association models generally out-performed fire history models in terms of predictive capacity. For two species, habitat association models provided good discrimination capacity even though the species showed geographically varying post-fire responses. Our results support the dynamic vegetation hypothesis, that spatial variation in relationships between fire and vegetation structure results in regional variation in fauna–fire relationships. These observations explain how the widely recognised ‘habitat accommodation’ model of animal succession can be conceptually accurate yet predictively weak.