Predator-prey models have traditionally been built on the assumption that neither predators nor prey are capable of learning or adapting behavior based on past interactions. As such, these models typically account for predation by forcing prey to experience a fixed level of predation risk. In order to develop species specific and more realistic models, the enemy-victim interaction behaviors of both predators and prey need to be studied. We quantified selection patterns for 2 fine scale behaviors of a recovering wolf population in Yellowstone National Park, Wyoming. Wolf spatial data were collected between November and May from 1998-1999 until 2001-2002. Over four winters; 244 aerial locations, 522 ground based telemetry locations, 1,287 km of movement data from snow-tracking and the locations of 279 wolf kill sites were recorded. There was evidence that elk (Cervus elaphus) and bison (Bison bison) densities had a weak effect on the sites where wolves traveled and made kills. Wolf movements showed a strong selection for geothermal areas, meadows and areas in close proximity to edge. Proximity to edge and habitat class also had a strong influence on the locations where elk were most vulnerable to predation. There was little evidence that wolf kill sites differed from the places where wolves traveled, indicating that wolves selected to travel in areas where they were most successful at killing elk. Our results indicate that prey are more vulnerable to predators under certain conditions, predators are capable of selecting for these conditions, and as such, the notion of a fixed predation risk is largely inadequate.