Some methods are evaluated and extended to estimate roughness length and zero plane displacement height for atmospheric flow over arrays of obstacles, typically buildings. It appears that the method proposed by Bottema, with an extension to account for low density obstacle arrays, performs best. Procedures are proposed to represent irregular obstacle arrangements by a representative regular array to which Bottema's method can be applied. It is shown that this can be done without loss of accuracy, in general, roughness length can be predicted within a factor of two in more than 74% of the cases (95% reliability estimate). The methods proposed by Lettau and Raupach have been included in the evaluation. Lettau's model, which only requires input on the frontal area density, predicts roughness length unbiassed for frontal area densities up to 0.3, but predictions will be within a factor of two in more than 59% of the cases only (95% reliability estimate).