The probability density function (pdf) of land-surface wind speeds is characterised using a global network of observations. Daytime surface wind speeds are shown to be broadly consistent with the Weibull distribution, while those at night are generally more positively skewed than the corresponding Weibull distribution (particularly in summer). In the middle latitudes, these strongly positive skewnesses are shown to be generally associated with conditions of strong surface stability and weak lower-tropospheric wind shear. Long-term tower observations from Cabauw, Netherlands demonstrate that lower-tropospheric wind speeds become more positively skewed than the corresponding Weibull distribution only in the shallow (~50 m) nocturnal boundary layer. This skewness is associated with two populations of night-time winds: one strongly stably stratified with strong wind shear, and the other weakly stably or unstably stratified with weak wind shear. Using an idealised two-layer model of the boundary layer momentum budget, it is shown that the observed variability of the daytime and nighttime surface wind speeds can be accounted for through a stochastic representation of intermittent turbulent mixing at the nocturnal boundary layer inversion.