Context. Feral cats are a major cause of mammal declines and extinctions in Australia. However, cats are elusive and obtaining reliable ecological data is challenging. Although camera traps are increasingly being used to study feral cats, their successful use in northern Australia has been limited.

Aims. We evaluated the efficacy of camera-trap sampling designs for detecting cats in the tropical savanna of northern Australia. We aimed to develop a camera-trapping method that would yield detection probabilities adequate for precise occupancy estimates.

Methods. First, we assessed the influence of two micro-habitat placements and three lure types on camera-trap detection rates of feral cats. Second, using multiple camera traps at each site, we examined the relationship between sampling effort and detection probability by using a multi-method occupancy model.

Key results. We found no significant difference in detection rates of feral cats using a variety of lures and micro-habitat placement. The mean probability of detecting a cat on one camera during one week of sampling was very low (p = 0.15) and had high uncertainty. However, the probability of detecting a cat on at least one of five cameras deployed concurrently on a site was 48% higher (p = 0.22) and had a greater precision.

Conclusions. The sampling effort required to achieve detection rates adequate to infer occupancy of feral cats by camera trap is considerably higher in northern Australia than has been observed elsewhere in Australia. Adequate detection of feral cats in the tropical savanna of northern Australia will necessitate inclusion of more camera traps and a longer survey duration.

Implications. Sampling designs using camera traps need to be rigorously trialled and assessed to optimise detection of the target species for different Australian biomes. A standard approach is suggested for detecting feral cats in northern Australian savannas.