Despite the widespread use of DNA mark–recapture for estimation of grizzly bear (Ursus arctos) population size, there have been no designed experiments of DNA sampling strategies. We designed a large-scale study (8,820 km²) in the foothills of Alberta, Canada, to test sampling strategies associated with the hair snag DNA method. The main sampling method for this project used a traditional design in which bait sites were moved within 180 7 x 7 km grid cells for 4 2-week sampling sessions in the spring of 2004. However, we also tested other strategies concurrently with the traditional design. We sampled fixed sites within each cell to test the utility of moving sites compared to the less-expensive method of not moving sites. We also placed a second, lower strand of barbed wire on bait sites to see if this could identify cubs, which are not typically sampled by the usual knee-height strand of barbed wire. We compared summary statistics, capture probability variation, population estimates, and the precision of population estimates for each design. The moved-sites designs captured more bears each session, captured more individual bears (especially females), and displayed population estimates that were 15–25% higher for females. Estimates for males were similar between designs. These results suggest that the moved-sites designs were more efficient in sampling the entire population at the 7 x 7 km grid cell size. These results highlight the need for all bears to have adequate trap encounter opportunities to ensure unbiased estimates. It also demonstrates the utility of collecting enhanced data sets to test and optimize DNA sampling strategies.