Sea level has been estimated for the last 108 million years through backstripping of corehole data from the New Jersey and Delaware Coastal Plains. Inherent errors due to this method of calculating sea level are discussed, including uncertainties in ages, depth of deposition and the model used for tectonic subsidence. Problems arising from the two-dimensional aspects of subsidence and response to sediment loads are also addressed. The rates and magnitudes of sea-level change are consistent with at least ephemeral ice sheets throughout the studied interval. Million-year sea-level cycles are, for the most part, consistent within the study area suggesting that they may be eustatic in origin. This conclusion is corroborated by correlation between sequence boundaries and unconformities in New Zealand. The resulting long-term curve suggests that sea level ranged from about 75–110 m in the Late Cretaceous, reached a maximum of about 150 m in the Early Eocene and fell to zero in the Miocene. The Late Cretaceous long-term (107 years) magnitude is about 100–150 m less than sea level predicted from ocean volume. This discrepancy can be reconciled by assuming that dynamic topography in New Jersey was driven by North America overriding the subducted Farallon plate. However, geodynamic models of this effect do not resolve the problem in that they require Eocene sea level to be significantly higher in the New Jersey region than the global average.