The over-estimation of mean wind speed in a variable wind is investigated for Sheppard and Meteorological Office patterns of the 3-cup anemometer. Comparison of Schrenk's theory with observation shows satisfactory agreement for hemispherical cups. Conical cups over-estimate mean wind speed to a much smaller extent than hemispherical cups and are to be preferred therefore when unsteady winds, e.g. natural winds, are to be measured. For a given cup form over-estimation appears to be a minimum for minimum mass per unit area of cup; other parameters such as cup diameter and arm length are of secondary importance.