Relative sea level (RSL) reconstructions from paleo records are often the most valuable data set for testing and constraining glacial isostatic adjustment (GIA) models. In some regions, the amplitude and rate of the reconstructed RSL changes are large making the data difficult to fit. Arguably, the reconstructed RSL curve from Maine, USA, is the classic example of such a data set with peak values at about 120 m dropping to a low stand of around −40 m within a few thousand years. To our knowledge, no GIA model has captured these extreme variations and the record has been somewhat neglected by the GIA modelling community. Here we critically assess and present a revised pre-10 ka RSL data base for this region and combine it with two recent Holocene compilations. Based on our assessment, we conclude that the large and rapid changes, inferred in previous studies, are robust. To determine if a successful model fit can be found, we consider a GIA parameter set comprising five ice models and 440 spherically-symmetric, Earth viscosity models assuming a Maxwell rheology. Results show that none of the model parameter sets produce a good fit to the data set. The rapid RSL fall and lowstand can be simulated by Earth models with low values of upper mantle viscosity (∼5 × 1019 Pas). However, such low viscosity values result in a large mid-Holocene highstand (order 10 m) while the observations indicate a monotonic RSL rise throughout this interval. Our results indicate that Earth models able to simulate time-dependent viscosity (i.e., those that include transient and/or non-Newtonian deformation) might be required to fit the Maine RSL data set.