Various models for calculating the effective or area-averaged roughness length zoe have been tested for a partly forested area. Three types of model are considered: the tile approach for very large scales of inhomogeneity (> 20 km), drag models for very small scales (up to 1 km), and surface-layer methods such as the blending-height method for intermediate scales. Over partly forested areas, where both pressure effects and roughness sub-layer effects may become significant, small-scale models are expected be the most suitable type of model. The various model types were tested against new experimental data that were obtained over the partly forested Sherwood Forest area (UK). The best fit with the data was obtained with the blending-height method, rather than with the different small-scale models. This is remarkable as the surface-layer assumptions of the blending-height method were clearly violated: the calculated blending height was 7 m, as compared to the mean tree height of 20 m. Of the small-scale models, a sparse-canopy approach compared poorly with the experimental data. The drag models overestimated the area-averaged roughness to a lesser degree, but a major point of concern remains the choice of the model parameters. Therefore, suggestions are made for an improved choice of these parameters.