The aim of this work is to evaluate the performance of two popular k − ϵ turbulence closure schemes for atmospheric boundary layer (ABL) flow over hills and valleys and to investigate the effect of using ABL-modified model constants. The standard k − ϵ and the RNG k − ϵ models are used to simulate flow over the two-dimensional analytical shapes from the RUSHIL and RUSHVAL wind tunnel experiments. Furthermore, the mean turbulent flow over the real complex terrain of Blashaval hill is simulated and the results verified with a data set of full-scale measurements. In general, all models yield similar results. However, use of ABL-modified constants in both models tends to decrease the predicted velocity and increase the predicted turbulent kinetic energy. Copyright © 2010 John Wiley & Sons, Ltd.