Spatial heterogeneity in predation risk can ameliorate impacts on prey populations, particularly for prey of generalists. Spatially heterogeneous risk implies the existence of refugia, and the spatial scale of those refugia and their persistence over time affect whether prey can avoid predation by aggregating therein. Our objective was to quantify the magnitude, spatial scale, and temporal persistence of heterogeneity in risk of predation by white-footed mice (Peromyscus leucopus), an abundant generalist predator of gypsy moths (Lymantria dispar) and songbirds. We used track plates to measure white-footed mouse activity at > 170 trees in each of three forest plots in upstate New York during summers of 2003-2005. We quantified the mean and coefficient of variation of track activity among trees by fitting the beta-binomial distribution to data from each plot and study period. We measured temporal persistence by disattenuated autocorrelation, and spatial scale by fitting exponential variograms. Mice were much less abundant in 2005 than the other two years, leading to lower overall track activity but higher coefficient of variation among trees. Mouse track activity at individual trees was positively autocorrelated between monthly study periods in 2003 and 2004, and even between the two years, whereas temporal autocorrelation in 2005 was much weaker. Track activity showed positive spatial autocorrelation over lag distances from approximately 30 to > 1000 m. These findings indicate that mouse activity, and hence risk to their prey, varies substantially in space at spatial and temporal scales that appear responsive to mouse population dynamics. The spatial scale and temporal persistence of that variation imply that prey may benefit from returning to, or failing to disperse from, refugia.