Crossover sea level differences from almost 5 years of TOPEX/POSEIDON data are regressed against corresponding differences in atmospheric pressure (-Δpa), and the regression coefficient (α) is examined for deviations from the value of ~1 cm/mbar expected under a pure inverted barometer (IB) signal. Only crossovers within each 10-day repeat cycle are used. We focus on variability at the shortest periods where non-IB response is more likely. Results indicate a marked tendency to have α<1 cm/mbar, with values in the general range of 0.8--1 cm/mbar in middle and high latitudes and 0.4--0.8 cm/mbar in the tropics. Effects of errors in Δpa and altimeter data seem small to explain all of the observed IB deviations, which imply then positive correlations between Δpa and dynamic sea level Δ&eegr;' (sea level adjusted for an IB signal). A simple constant-density ocean model is used to help interpret the regression results. Only effects of winds and pa on Δ&eegr;' are considered. Model-based α estimates agree qualitatively with the data estimates. On the basis of the model results, wind-driven Δ&eegr;' signals contribute importantly to the spatial variability of α observed at middle and high latitudes. This is particularly evident in the Southern Ocean, where strongest wind-driven effects on α coincide with regions of anomalous ambient potential vorticity gradients and different vorticity dynamics. In contrast, the observed decrease in α toward low latitudes is due to a true dynamic response to pa, mostly at periods shorter than 10 days. The stronger non-IB signals suggest a response closer to resonance and may be due to the richer resonance spectrum in the tropics. The non-IB signals are largely remotely driven and therefore weakly correlated with local pa. Thus regression results tend to underestimate the importance of the nonisostatic response in the tropics but also at higher latitudes. ¿ 1999 American Geophysical Union