The fencing of protected areas is increasing worldwide. However, the implementation of fences for conservation has outpaced scientific assessment of their effectiveness, non-target impacts, and long-term costs. We assessed landscape predictors of fence crossing sites and employed camera traps over a one-year period to investigate wildlife responses to a conservation fence around Lake Nakuru National Park, Kenya. Specifically, we measured the impact of the fence on wild mammal movement, and the temporal impacts of fence maintenance on wildlife crossings and behavior. Cameras captured more than 65,000 detections of animals approaching fences, with 3626 observed crossings over 2818 trap nights at 19 sites. Using these data, we developed a guide to classifying fence-specific mammal behaviors. Thirty-eight wild mammal species approached known weak points in the fence, and 27 species were recorded crossing the fence. No single environmental variable predicted detection or fence crossing points for all species, but seasonality, human activity, habitat visibility, and proximity to an adjacent protected area were each correlated with species-specific crossing locations. Additionally, breaches of repaired fence-crossing locations occurred within days of maintenance. We conclude that popular, 'one-size-fitsall', conservation fence designs may be ineffective and costly for restraining movement of many wildlife species. We recommend that those deploying conservation fences start with clearly articulated management goals, that fence maintenance be informed by taxa-specific tendencies to breach fences, and that managers consider the strategic creation of wildlife corridors, overpasses, or ungulate-proof fences to link fenced protected areas with surrounding habitat.