The uncertainty in hypocenters and origin times depends on measurement error (the wrong onset is picked) and error in the travel-time tables (and thus earth model) used. The errors in the tables comprise a baseline shift (the average difference over all stations is not zero) and the residuals about the baseline. The residuals are usually referred to as model error. Baseline error only affects origin time and so is usually ignored. It is model error that can result in epicenter error. A priori variances of the model and measurement error are usually used to estimate the uncertainty on epicenters. Few estimates of these variances have been published. Here the two variances are estimated, relative to International Association for Seismology and the Physics of the Earth's Interior (iaspei) 91, for the P times from explosions at the Nevada Test Site at stations at regional distances. The analysis shows that at a large signal-to-noise ratio (snr) the variance of the measurement error is 0.01 sec2. The measurement error increases as snr decreases. Further, the travel times appear to increase as snr decreases. Model error has a formal variance of up to 1.9 sec2, but this variance is irrelevant to assessing the uncertainty in epicenter estimates. It is systematic bias (if any) caused by model error that contributes to epicenter uncertainty. Without knowing the bias it is only possible to estimate the precision of an epicenter and this depends on the measurement error. The analysis gives estimates of model error for each source-to-station path, and these path effects can be used to correct the travel times to give a revised model. With correction for path effects the estimated uncertainty in the epicenters becomes a measure of the accuracy of a location. The results presented here show that when estimating these effects, care must be taken to ensure that variations in snr do not bias the estimates.