We used isodars to analyse habitat-dependent population regulation by long-nosed bandicoots Perameles nasuta during an irruption and subsequent population crash in three habitats (heath, woodland and forest) at Booderee National Park, south-eastern Australia. Specifically, we aimed to see whether patterns of habitat-dependent population regulation matched a priori estimates of quantitative and qualitative differences between habitats. We also tested if habitat preference changed between the increasing and decreasing phase of the irruption as predicted by the reciprocating dispersal theory. Quantitative differences in habitat quality were indexed by the relative abundance of the main food of long-nosed bandicoots (terrestrial invertebrates), while qualitative differences were indexed by the availability of refuge from predation (vegetation understorey density). One index of fitness, body weight, was highest in forest, and lowest in heath, suggesting an ideal despotic model of habitat selection. Over the entire course of the irruption, there was density-dependent habitat selection with forest and woodland both quantitatively superior to heath. This reflected the overall abundance of invertebrates with highest abundance in woodland and forest and less in heath. Isodar analysis also revealed that although forest was quantitatively better than heath and equivalent to woodland it was qualitatively poorer than either habitat. Heath had a higher density of understorey than woodland and woodland having a higher density of understorey than forest giving crossover population regulation. When the increasing and declining phase of the irruption were analysed separately, no habitat was quantitatively superior to any other during either phase. The lack of switching in preference between habitats from the increasing to the declining phase of the irruption and the virtual absence of any dispersal by adults, does not support the reciprocating dispersal hypothesis.