The Haskell <1935> value of 1021 Pa s for mantle viscosity is a classic and enduring constraint on the rheology of the Earth's interior. We revisit this inference using spherically symmetric, self-gravitating, Maxwell viscoelastic Earth models. Our inference is based on both forward and inverse analyses of decay times associated with uplift at two sites considered by Haskell, Angerman River, Sweden, and Oslo, Norway, rather than the raw relative sea level (RSL) data at these sites. We demonstrate that predictions of the decay time associated with the Angerman River data are insensitive to variations in both the late Pleistocene ice load history and the lithospheric thickness of the Earth model (the predictions at Oslo are sensitive to both these inputs), and hence decay times at this site provide a remarkably robust constraint on mantle viscosity. We derive a constraint on the ''average'' viscosity of the mantle of 0.65--1.10¿1021 Pa s, where the ''average'' resolved by the data encompasses a region which extends from the base of the lithosphere to a depth of near 1400 km. This indicates that many previous analyses which have invoked the Haskell value of 1021 Pa s as a constraint on the average upper mantle (i.e., above 670 km depth) viscosity alone have misinterpreted the resolving power of the inference. Furthermore, our analysis indicates that a number of apparently contradictory inferences of viscosity based on Fennoscandian data satisfy the new, rigorous, interpretation of the Haskell constraint. Finally, we demonstrate how the ambiguity in the upper mantle/lower mantle viscosity contrast associated with the Haskell ''average'' may be reduced by invoking decay time constraints estimated from RSL curves in Hudson Bay. A preliminary inversion of decay times at Angerman River and Richmond Gulf (in Hudson Bay) suggests a contrast of approximately an order of magnitude between the average viscosities of the upper mantle and the top 1000 km of the lower mantle; however, a conclusive analysis in this regard must await the determination of consistent decay time estimates for the Hudson Bay region. ¿ American Geophysical Union 1996