Abstract. The two commonly used statistical measures of the air-sea heat flux, the sampling and classical means, have been compared using hourly reports over a 7-year-period from a weather ship stationed in the NE Atlantic. The sampling mean is the average over all flux estimates in a given period, where individual flux estimates are determined from ship reports of meteorological variables using the well-known bulk formulae. The classical mean is the flux derived by substituting period-averaged values for each of the meteorological variables into the bulk formula (where the averaging period employed is the same as that over which the fluxes are to be determined). Monthly sampling and classical means are calculated for the latent and sensible heat fluxes. The monthly classical mean latent heat flux is found to overestimate the sampling mean by an amount which increases from 1–2 W m–2 in summer to 7 W m–2 in winter, on average, over the 7-year-period. In a given winter month, the excess may be as great as 15 W m–2, which represents about 10% of the latent heat flux. For the sensible heat flux, any seasonal variation between the two means is of the order of 1 W m–2 and is not significant compared to the interannual variation. The discrepancy between the two means for the latent heat flux is shown to arise primarily from a negative correlation between the wind speed and sea-air humidity difference, the effects of which are implicitly included in the sampling method but not in the classical. The influence of the dominant weather conditions on the sign and magnitude of this correlation are explored, and the large negative values that it takes in winter are found to depend on the typical track of the mid-latitude depressions with respect to the position sampled. In conclusion, it is suggested that sampling means should be employed where possible in future climatological studies.