Estimates of population size are frequently used in conservation. Volunteer-conducted surveys are often the only source of information available, but their reliability is unclear. We compare data from a weakly structured national bird atlas collected by volunteer surveyors free to choose where and when to visit with data from an independent suite of monitoring surveys that used a stratified sampling design. We focus on the Mount Lofty Ranges, South Australia, a region that has lost most of its native vegetation. Both datasets comprise several thousand 20-min 2-ha searches carried out between 1999 and 2007. The atlas dataset reported more species, and covered habitats more comprehensively, but showed greater variability in the temporal and spatial distribution of survey effort. However, after we restricted the atlas dataset to native eucalypt woodlands, reporting rates from the two schemes were very strongly correlated. The structured surveys tended to record more species that are normally detected by call and the unstructured surveys recorded more species using edges and open habitats. Minimum population estimates from the two datasets agreed very well. The strength of concordance depended on whether overflying birds were included, highlighting the importance of distinguishing such records in future surveys. We conclude that appropriate calibration using selected regional surveys, including surveys to estimate absolute densities, can enable volunteer-collected and weakly structured atlas data to be used to generate robust occupancy and minimum population estimates for many species at a regional scale.