The probability of extreme sea levels along the coasts has always been statistically estimated from the series of local observations. The inference is never conclusive, and an attempt is given here to improve the methods already used with reference to the area of the English Channel. The joint probability method (JPM) is the starting point: In most cases it underestimates the return times (or overestimates extreme levels at a fixed time). The proposed extension is based on a more careful use of observed extremes by fitting a coefficient Cc deduced from the data set, which requires that the maximum record height be in agreement with the return period of the record length. This correction calibrates the whole series of extreme estimations to the observed maximum. Likewise an attempt to roughly explain this correction is given that explores the tide–surge interaction and seasonal dependence. The parameters are specifically computed for 15 tide-gauge stations, and the comparison is extended to other known methods, like the Gumbel one (in most cases overestimating the levels) and GEV simulations (which appear much better). Finally extreme levels with estimated return times of 10, 50, and 100 years, respectively, are proposed for each site, and a test for validity was performed by splitting certain long records into small samples, thus checking the spread of the results.