Camera trapping has recently been introduced as an unbiased and practical method for monitoring tiger abundance. In a high density area in the Royal Bardia National Park in lowland Nepal, we tested this method by trapping very intensively within a 25 km2 area to determine the true number of animals in that area. We then tested the effect of study design by sub-sampling the data set using varying distances between trap stations and by reducing the number of trapping nights at each station. We compared these numbers with the density estimates generated by the capture–recapture models of the program CAPTURE. Both distance between traps and trapping duration greatly influenced the results. For example, increasing the inter-trap distance from 1 to 2.1 km and reducing the trapping duration per station from 15 to 10 nights reduced the number of tigers captured by 25%. A significant decrease in trapping rates during successive 5-night periods suggested that our tigers became trap-shy, probably because of the photo flash and because they detected the camera traps from cues from impression pads 50 m from the traps. A significant behavioural response was also confirmed by the program CAPTURE. The best capture–recapture model selected by the computer program (Mbh) gave precise estimates from data collected by the initial 1 km spacing of traps. However, when we omitted data from half the number of traps, thus decreasing the sampling effort to a more realistic level for monitoring purposes, the program CAPTURE underestimated the true number of tigers. Most probably, this was due to a combination of trap shyness and the way the study was designed. Within larger protected areas, total count from intensive, stratified subsampling is suggested as a complementary technique to the capture–recapture method, since it circumvents the problem of trap shyness.