Error characteristics of a mirror loss measurement system using an exponential decay method were studied, with the two samples having about 200-ppm-loss and 30-ppm-loss, respectively. In order to minimize the decay signal deviation from an exponential curve due to cavity length fluctuation, a data average method was tried. The data average method significantly improved the exponential curve fitting error of the decay signal, so that for a 6 decay signal data average the loss measurement error was reduced by about 2.4 times for the 200-ppm-loss mirror and 1.3 times for the 30-ppm-loss mirror compared with a single shot measurement. Day-to-day mirror loss repeatability error for the two samples was investigated. The repeatability error was measured to be about 5% for the 200-ppm-loss mirror and about 26.4% for the 30-ppm-loss mirror. Low decay signal average effect and high repeatability error in the low loss mirror measurement were explained with non-uniform spatial loss distribution of the sample and contamination from the environment, in addition to the error sources of the mirror loss measurement system itself. The influence of cavity length fluctuation and cavity length measurement error on the mirror loss measurement system performance was theoretically calculated. It confirmed that the requirement for the cavity length parameters was not so strict in the mirror loss measurement system of several ppm resolution. ution.