It is well-known that the power spectral density of a random process may be estimated from its autocovariance function. Here I report on efforts to compute the spatially-averaged frequency spectrum and autocovariance of sea level from dual-satellite crossover data. By averaging over all crossover locations, and by using time-lagged sea level differences binned with hourly resolution, combined Jason and Cryosat-2 data yield power spectral density estimates with a Nyquist frequency of $0.5$ cycles-per-hour. While the same sea surface is observed by the Jason missions and Cryosat-2, their data are contaminated by independent realizations of measurement noise, so some aspects of the autocovariance are interpreted differently than in conventional time series analysis. Because the crossover locations are not randomly distributed in space and time, there are artifacts related to the aliasing of spatial signals that appear in the frequency domain. Nonetheless, the resulting power spectral estimates clearly exhibit unaliased peaks due to high-frequency sea level above the 1/20-day Nyquist frequency associated with Jason. I will provide examples of these spectral estimates and show how they may be used to learn about non-phase-locked baroclinic tides.