Accurate estimates of breeding population size are essential for detecting change and guiding conservation management and sustainable use. In New Zealand, the grey-faced petrel (Pterodroma macroptera gouldi) now breeds predominantly on offshore islands, but is also subject to customary harvest of chicks by northern tribes of Maori (New Zealand's indigenous peoples). We used island-wide surveys of 3186 breeding burrows, corrected for detection error, and associated habitat variables on two island systems (Ruamaahua Islands and Moutohora), combined with data from geographic information systems, to build hierarchical Bayesian models to predict the distribution and abundance of breeding pairs. Burrow densities increased with elevation on all islands and, on Moutohora, were lesser in gullies and on terraces. On the Ruamaahua Islands, burrow densities were associated positively with deeper soils and forests dominated by pōhutukawa (Metrosideros excelsa), and negatively with taller forest, denser canopy cover, and northern and western aspects. Predicted breeding-pair densities for each island group provided breeding-pair population estimates of 72,410 (95% credible interval 14,280-138,400) across the Ruamaahua Islands and 69,330 (10,590-128,300) on Moutohora. Estimating burrow densities using habitat characteristics provided greater precision and accuracy than simpler models that extrapolate sampling data over larger areas. The methods used to estimate population size are applicable to other cryptic seabird species, especially those that live or breed in burrows. It is important to establish accurate baseline estimates of the populations of common seabird species against which to determine responses to perturbations and management interventions.