The distribution of hollow-using fauna is frequently related to forest type. If hollow occurrence varies with forest type, the use of generic prescriptions for managing the hollow resource may be inappropriate. This study examined the relationship between the occurrence and abundance of tree hollows and site and tree attributes in wet and dry Eucalyptus obliqua forest in Tasmania, Australia. A total of 388 trees at 39 sites were examined before and after being felled. Evidence of invertebrate damage contributing to hollow formation was more common in wet forest than either dry or damp forest. Evidence of hollow formation by fire and fungi was more prevalent in drier than wetter forest. There was no difference among forest types in the proportion of hollows showing evidence of limb breakage. Examination of the site and tree factors related to hollow presence and abundance was done using Classification Trees, Random Forests and generalised linear models. The variables found to be most practical for predicting hollow occurrence were the number of hollows observed before the tree was felled, tree diameter and the amount of dead wood in the canopy. Tree age, an important variable to consider when planning harvest rotations, was strongly associated with hollow presence but had less bearing on hollow abundance. The size of hollows that were found in a tree was related to greater senescence. Although significantly more hollows of all sizes were found in wet forest than either dry or damp forest, the age at which trees began to produce hollows was similar among the different forest types. The models predicted that trees needed to be at least 100 years old before they are likely to contain a hollow and that medium-sized hollows rarely occur in trees younger than 140 years old. Trees grow slightly more slowly in dry forest than wet which means that smaller diameter trees were found to contain hollows in dry forest than wet and that even the largest trees may not contain large hollows in wet forest. Despite this, no differentiation between the forest types was found in the predictive models. It is recommended that trees retained for faunal habitat be at least 100 cm in diameter.