Examining the predictive capability of statistical models with data independent from that used to derive the model is a vital step in the iterative procedure of assessing model performance. I derived logistic regression models of the habitat use of the rufous treecreeper (Climacteris rufa) at two spatial scales: woodland (territory selection model) and territory (nest-site selection model). The performance of these models was assessed in relation to the original data collected and validated with new, independent data. When applied to the original data, the territory model had a high predictive capability correctly classifying 90% of sites (n=100) that were either occupied or unoccupied by treecreepers. Correct classification rate was reduced to 70% (n=50) when the model was applied to the validation data. Model performance was generally robust when probability of occurrence values for the species were varied. In contrast, the nest-site model had lower predictive capabilities correctly classifying between 66 and 68% of sites, and performed relatively poorly when probability values were varied. The performance of the models differed slightly between the original and validation data, and substantially between the spatial scales examined. Territory use by rufous treecreepers could be predicted with some confidence indicating that the territory model may be a useful tool for habitat management. Nest-site use could not be predicted with confidence probably as a result of the high abundance of suitable, but unused, nest sites in the study area.