Knowledge of the probability distributions of surface wind speeds (SWS) is essential for surface flux estimation, wind power estimation, and wind risk assessments. The two-parameter Weibull distribution is the most widely used empirical distribution for SWS. This study considers the probability density function (PDF) of 3-hourly observations from 720 weather stations over North America for the period 1979–1999. The PDF of SWS is classified by season, time of day, and land surface type. The Weibull PDF is characterized by a particular relationship between the mean, standard deviation, and skewness. While the moments of the observed daytime SWS PDF are found to collapse around this Weibull relationship, the observed nighttime PDF has a broader range of values and is significantly more skewed than the Weibull PDF over rough surfaces. An idealized model shows that SWS skewness has a much greater rate of change with both the mean and standard deviation of surface buoyancy flux under conditions of stable stratification than that of unstable stratification. This result suggests that surface buoyancy flux plays an important role in generating diurnal variation of SWS PDF. Two global reanalyses products (ERA-40 and NCEP-NCAR) and three regional climate models (RCMs) (Rossby Centre Atmospheric Model version 3 (RCA3), limited area version of Global Environmental Multiscale Model (GEM-LAM), and Canadian Regional Climate Model, version 4 (CRCM4)) all have a less skewed nighttime PDF and a more narrow range of the normal wind speed during day and night. Among them, two of the RCMs capture the observed SWS differences across different land cover types, and only one of the RCMs produces the observed seasonal peak of SWS PDF.