Domestic dogs are the most abundant carnivores worldwide, primarily due to human support. Food and other subsidies to dogs do not necessarily prevent dog predation on wildlife, particularly where dogs are allowed to range freely. Dog impacts on wildlife are suspected to be significant, yet the nature of dog-wildlife interactions is not fully understood. We tested the hypothesis that the distribution of dogs can significantly influence the space use of potential prey, and that both lethal and non-lethal mechanisms may underlie this interaction. If this is true, then we predicted that (1) evidence of predation and harassment by dogs should be evident where prey and dog activities overlap and (2) potential prey should be less frequent in areas where the probability of dog presence is high. To test these predictions we conducted two related studies. (1) We interviewed dog owners to estimate the probability of dog attack on pudu (Pudu puda), a globally vulnerable deer, and the lethality of these attacks. (2) We conducted a camera-trap survey documenting the landscape-scale distribution of pudu and dogs. Interviews showed that both the probability of dog attack on pudu (>85%) and the lethality of such attacks was high (50%). In occupancy models applied to the camera-trap data, the variable that best explained the distribution of pudus was the probability of dog presence. We tested three alternative explanations for the negative association between pudus and dogs that were not supported. Our findings suggest that dogs are efficient at chasing pudu they detect and that both predation and non-lethal (avoidance) consequences of harassment may be shaping the distribution of pudu. This work brings into focus important mechanisms underlying the threats of domestic dogs to endangered prey. (C) 2012 Elsevier Ltd. All rights reserved.