Extreme sea levels along the densely monitored coasts of the North Atlantic Ocean and the Gulf of Mexico have been investigated using high frequency tide gauge measurements in the GESLA-2 data set (www.gesla.org). Our results, based on non-tidal residuals and skew surges in records since 1960, confirm that mean sea level (MSL) is a major, but not a unique, driver of extremes. Regionally-coherent linear trends and correlations with large scale climate patterns are found in extreme events, even after the removal of MSL. A similar conclusion, that MSL is a major but not the only driver of extremes, comes from a small number of long records starting in the mid-19th century. The records show slight increases in the intensity of extreme episodes at centennial time scales, together with multi-decadal variability unrelated to MSL. Objective statistical criteria have been used to investigate whether extreme sea level distributions are stationary or not, resulting in non-stationarity being favoured in many records, with or without accounting for changes in MSL. Extremes have been found to favour a non-Gumbel behaviour at many locations, with implications for the accuracy of return levels for coastal engineering.