Context. Low-volume control agents based on new biotechnologies are likely to be point-delivered to wildlife populations using devices such as bait stations. However, data and theory to underpin the development of strategies for such a use of bait stations are lacking.

Aims. In a large-scale replicated field trial of brushtail possum (Trichosurus vulpecula) populations, we estimated the levels of coverage achieved with bait-station grids at three densities (0.2, 0.6 and 1.0 ha^–1), with and without pre-feeding, to test the predictions of an existing model of bait-station delivery based on relationships derived from leghold trapping.

Methods. We first marked possums using conventional capture–mark–recapture techniques, and then estimated population coverage by recovering animals poisoned by encapsulated cyanide delivered by the bait stations.

Key results. The two key model predictions, that 90% population coverage could be achieved at 0.6 bait stations per hectare and that pre-feeding was not required to achieve this level, were not upheld by the field trials. Instead, there was a poor fit between model predictions and observed levels of population coverage. An investigation of key model assumptions demonstrated issues with both model structure and parameterisation.

Conclusions. Neither previously documented relationships for possum interactions with leghold traps, nor correlations between possum interaction rates with such devices and bait stations, are generically applicable across all populations.

Implications. While the existing model of bait-station delivery to possums fits the data for some field operations and trials, it is not applicable to all populations. Incorporating habitat and seasonal effects on possum home-range behaviour could potentially improve model prediction. Our results also have implications for the accuracy of index-based methods of possum population monitoring that use leghold traps, such as the Residual Trap Catch Index.