The stabilization of exposure dose repeatability is one of the key factors in reducing the critical dimension variation in synchrotron radiation lithography. We have evaluated and improved the exposure dose repeatability of the full field exposure system, which employs a cylindrically convex mirror. The beam size spreads with the stored current in the storage ring, which causes the major exposure dose errors. The maximum exposure dose error is evaluated to be +/-4.0% according to the ray trace simulation, and it is expected to reduce the error to +/-0.6% using exposure time compensation. We have evaluated the improvement of the exposure dose repeatability experimentally by using the stabilized resist process. Critical dimension variations of less than +/-2 nm between six wafers exposed over several days were obtained. This result indicates that the practical exposure dose error is almost consistent with the calculated value, having a variation of less than +/-1.0%.