The design of sampling schemes affects the results of biodiversity inventories. As an approach for quantifying the implications of such effects, we compared data on spider communities sampled in a beech-dominated forest floor habitat by 1) a regular grid of pitfall traps (systematic design) and 2) an expert-based distribution of traps (stratified design). We tested whether the two designs would lead to similar conclusions about the diversity and composition of ground-dwelling spider communities. Estimates of species richness, rarefied species richness and activity density calculated per trap were significantly higher in the stratified than in the systematic design. The community composition based on the presence or absence of sampled species or based on log-transformed activity densities differed significantly. Most of the dissimilarity between the community estimates of the two designs was attributable to three species, with Pardosa saltans Töpfer-Hofmann 2000 being more common in traps of the stratified design and Tenuiphantes zimmermanni (Bertkau 1890) and Walckenaeria cuspidata Blackwall 1833 being more frequently observed in traps of the systematic design. Our study suggests that a stratified sampling design is better suited for inventory surveys of spider communities of forest-floor habitats, as trap locations of this design reflect specific habitat needs. It is important to note that inventories are a major field for the application of such designs and that greater care is needed for the application of inferential statistics. For example, the non-randomness that is caused by expert selection of sampling sites may violate fundamental assumptions of simple linear models.