We characterize the accuracy of a wildlife localization system that is based on the reverse-GPS or time-of-arrival (TOA) principle, in which radio receivers at known locations collaborate to determine the location of a transmitter attached to a wild animal. We describe the system in detail and show that it produces accurate location estimates in real settings and over long periods of time. Localization errors of wild animals carrying our transmitters have standard deviation of about 5m and mean of 5-15m. If we restrict the system to operate in the center of the receiver's band-pass filters, the mean error drops to about 5m. We also show how to reliably quantify the error in each individual location estimate. In addition to the characterization of accuracy, we introduce three technical innovations in the system. First, a method to model the error of individual time-of- arrival measurements, enabling correct weighing of the data to estimate locations and allowing estimation of the covariance matrix of each location estimate. Second, extensive use of known- position beacon transmitters, to synchronize the clocks of receivers (radio-frequency TOA localization require accurate clocks), to characterize and continuously monitor the performance of the system, and to model arrival- time-estimation errors. Third, we estimate the covariance matrix of each localization.