Sea level variations relative to a 4‐year mean in the Indian Ocean north of 10°S are examined during 1993–1996 using both a numerical reduced gravity model with realistic coastline geometry and wind stress and sea level measurements from the TOPEX/Poseidon altimeters. The annual signal is found to be composed of propagating as well as nonpropagating features. The propagation speeds in the model and altimetry generally agree to within 25% or less. Complex empirical orthogonal function (CEOF) decomposition yields a separation between the annual and semiannual cycles (46 and 30% of the respective variance for the model, and 40 and 26% for the altimetric measurements, respectively). The propagation of these signals across the ocean basin is indicated by the spatial phase functions. Both temporal phase functions are steady for the annual cycle, though the amplitudes are modulated in time. The results for the semi‐annual cycles are similar, but the temporal phase functions are disrupted for ∼12 months starting in 1994. This may be due to an unusually strong monsoon during that time. The correlation between model sea level variation and those measured by altimetry is highly variable in both space and time. Low‐frequency filtering of the sea level anomalies, obtained by summing the two largest CEOF modes (the annual and semiannual cycles), improves the correlation. The filtered anomalies correlate in time as high as 0.9 in the western Arabian Sea and as high as 0.7 south of the equator and in the eastern Bay of Bengal. There are pockets of poor correlation (as low as −0.4) in the eastern Arabian Sea, central Bay of Bengal, and central equatorial region. These areas tend to contain recurring Rossby wave interactions as represented by the 1.5‐layer model. Each area is associated with a “phase nexus” (analogous to an amphidromic point in tidal theory) or a strong gradient of the model spatial phase functions. The spatial correlation between the filtered anomalies is typically 0.6 over much of the observation period but contains unexplained declines as low as 0.2 during a few months in bot