Observations of atmospheric flow over a change in surface roughness are reported. Both wind speed and turbulence characteristics were measured. Although the observation site departed from the ideal assumed in roughness change models, it was found that the predictions of `second-order closure' models are consistent with the observed flow