We characterize the error distribution of BATSE GRB locations by modeling the distribution of separations between BATSE locations and IPN annuli. We determine error model parameters by maximizing likelihood and rank the models by their Bayesian odds ratios. The best models have several systematic error terms. The simplest good model has a 1.9 degree systematic error with probability 73&percent; and 5.4 degrees with probability 27&percent;. A more complex model adds a dependence on the datatype used to derive the location. ©1998 American Institute of Physics.