We conducted the most intensive estimate of the endangered Lower Keys marsh rabbit (Sylvilagus palustris hefneri) metapopulation to date using pellet surveys and capture—recapture methodology. We livetrapped 83 rabbits, evaluated 5 closed population models, and selected the model that best represented the data. We considered the variation in behavioral response model the best model and correlated (r² = 0.913) its patch population estimates to patch pellet densities. From the prediction equation, we generated a range-wide metapopulation estimate of 317 rabbits, a western clade population of 257 rabbits, an eastern clade population of 25 rabbits, and translocated marsh rabbit populations of 35 and zero on Little Pine and Water keys, respectively. A subset of patches whose marsh rabbit subpopulations were last estimated in 1993 exhibited a 46% decline in abundance over 15 yr. Due to the low estimate of the eastern clade population, special effort should be initiated to avoid loss of genetic diversity. The prediction equation suffers from limited data at high pellet densities, patches with ≥5 pellets/m². Future studies should investigate if the slope of the regression is indeed near 1 by sampling patches across the range of pellet densities, especially those with ≥5 pellets/m². The equation provides managers a quick, efficient, and noninvasive method to estimate marsh rabbit abundance from pellet counts but the confidence of predicted rabbit densities from high pellet density patches is low.