Landscape metrics are commonly used indicators of ecological pattern and processes in ecological modelling. Numerous landscape metrics are available, making the selection of appropriate metrics a common challenge in model development. In this paper, we tested the performance of methods for preselecting sets of three landscape metrics for use in modelling species richness of six groups of organisms (woody plants, orchids, orthopterans, amphibians, reptiles, and small terrestrial birds) and overall species richness in a Mediterranean forest landscape. The tested methods included expert knowledge, decision tree analysis, principal component analysis, and principal component regression. They were compared with random choice and optimal sets, which were evaluated by testing all possible combinations of metrics. All pre-selection methods performed significantly worse than the optimal sets. The statistical approaches performed slightly better than random choice that in turn performed slightly better than sets derived by expert knowledge. We concluded that the process of selecting the most appropriate landscape metrics for modelling biodiversity is not trivial and that shortcuts to systematic evaluation of metrics should not be expected to identify appropriate indicators.