Analyses of long‐term changes of sea level often use monthly means for data. Beginning instead with an hourly series from which the predicted tides have been removed allows various errors in the record to be checked, since any departure from the predicted tides shows up strongly in the residual. Some recommended procedures are described, using as an example a record from Avila Beach, California, whose monthly means showed large excursions during 1969. The cause appears to have been clogging of the stilling well, but response analysis shows the nonlinear effects of this to have been too small to account for the observed shift in the mean. This shift appears to have come from the method of comparative readings used to establish the gauge zero, a procedure biased by lags in the gauge.