Camera traps are being increasingly used in biological surveys. One of the most common uses of camera trap data is the generation of species inventories and estimations of species richness. Many authors have advocated for increased camera trap-nights (long deployment times or more cameras in an array) to detect rare or wide-ranging species. However, in practice, the number of traps and the duration of surveys are constrained; a survey leader must make decisions about allocating the available cameras to sites. Here we investigate the effect of deployment time, camera array size and number of sites on detection of saxicoline mammal and varanid species obtained from surveys of discrete vegetation pockets in tropical Australia. This paper provides an analysis method for optimising decisions about how a limited number of cameras should be deployed across sites. We found that increasing the number of sites leads to larger species richness estimates in a shorter period. Increasing the number of cameras per site also leads to higher species richness estimates in a shorter time, but not to the same extent as increasing the number of sites. With fewer sites used or smaller arrays deployed at each site, a longer deployment duration is required, especially for rarer or wider-ranging species, or those not attracted to bait. Finally, we compared estimates of species richness generated by our camera trapping to those generated by live trapping at a subset of our sites, and found camera traps generated much larger estimates.