Extensive analyses of altimetrically determined sea height differences at crossovers have been used to assess the accuracy of the GEM-T2 geopotential. The orbits used were determined with GEM-T2 for Geosat in its 17-day Exact Repeat Mission (ERM) in 1986--1989 and ERS 1 in both its 3-day ERM in 1991--1992 and its 35-day ERM in 1992. The data examined are completely independent of the data used in GEM-T2's development though GEM-T2 had considerable use of Doppler tracking information on Geosat. The test of the radial accuracy of the ERS 1 orbit (98.5¿ inclination) is especially significant because it is not ''close'' to any other orbit well represented in GEM-T2. The assessment consists of a comparison of observed mean height differences at thousands of distinct geographic locations with error projections from the GEM-T2 covariance matrix which was estimated from other data sources. This first comprehensive, independent test of the purely radial accuracy of an orbit-geopotential model clearly shows that the covariant predictions for GEM-T2 are broadly reliable for this purpose. Thus, the agreement of crossover predictions and observations suggests that the total radial errors for these ERMs, due to only to GEM-T2 (but excluding the effects of initial state error) are about 23 cm for Geosat and 115 cm (rms) for ERS 1. However, there is little detailed agreement of measurements and predictions for ERS 1 and only partial agreement in detail for Geosat. Our 30,000 mean crossover discrepancies for Geosat (derived from ERM cycles 1--44) are also shown to reduce substantially the crossover height differences in cycles 45--61, almost exactly as predicted if these are the true GEM-T2 errors for this orbit. ¿ American Geophysical Union 1994