Inaccurate population data remain a major problem in the conservation of burrow-nesting Procellariiform seabirds. We evaluated the potential of predictive habitat modelling, in contrast with traditional surface-area based strategies, to estimate the population size of the endangered Cook’s petrel (Pterodroma cookii), breeding on one of two islands in New Zealand. Generalised linear models of burrow abundance using topographic data collected in the field, or extracted from Geographic Information System layers, were developed, and the final model applied to Geographic Information Systems-derived datasets in a population analysis. To compare this result with traditional census methods, population estimates were also calculated by averaging burrow densities across the total area available (simple model) and within vegetation zones (habitat-area model). Results of all models were scaled using rigorously collected burrow occupancy data. The final model, based upon altitude, slope, and distance from ridgelines, explained 31% of the variance in Cook’s petrel burrow abundance and showed an enhanced predictive fit compared with the two area-based models. This model indicated that annually 286,000 (95% confidence interval: 213,000–413,000) pairs of Cook’s petrel breed on Little Barrier Island, approximately six times that previously suggested. Population estimates based on simple and habitat-area models were less accurate, being 10% and 30% greater than the predictive model estimate, respectively. This study underlines the need for accurate population estimates for burrowing seabird taxa by presenting a major population revision for the Cook’s petrel. Predictive habitat models, employing rigorously collected data, offer an improvement on more traditional population census methodologies that are susceptible to scale-induced bias.