Viewing altimeter data only at the points where separate altimeter missions' ground tracks cross provides a method to observe long time period sea surface height (SSH) variations and avoids many of the problems inherent in combining separate altimeter data sets through an independently determined geoid. TOPEX/POSEIDON (T/P) data over the time period from January 1, 1993, to December 31, 1995, form a mean SSH that is used as a reference by other altimeter data sets. A least squares analysis of the mean T/P SSH determines the portion of the Geographically Correlated Orbit Error (GCOE) that may be observed through crossover differences and removes this portion of the GCOE. The analysis removes errors of 0.86 cm RMS at 1 cycle per orbit revolution (cpr) and indicates negligible errors at higher frequencies. After the GCOE removal, the accuracy of the T/P reference mean is better than 1 cm RMS as measured by crossover differences. The GCOE contained in the Geosat-Exact Repeat Mission (ERM) and ERS 1 data with orbit solutions using the Joint Gravity Model (JGM) 3 is evaluated through an adjustment to the T/P reference mean surface. The Geosat-ERM data indicate a bias of about 28 cm averaged over the globe, and the ERS 1 bias is 44 cm. The T/P data used here is not corrected for the oscillator drift correction error so that the actual bias is less by about 13 cm. Both the Geosat-ERM and ERS 1 GCOE are mainly 1 cpr. GCOE estimates at frequencies above 1 cpr indicate little actual orbit error but are more correlated to instrument correction errors (particularly water vapor). Simultaneous T/P and ERS 1 SSH anomalies to the T/P mean indicate good correlation. ¿ 1997 American Geophysical Union