Questions: Light availability at the forest floor affects many forest ecosystem processes, and is often quantified indirectly through easy-to-measure stand characteristics. We investigated how three such characteristics, basal area, canopy cover and canopy closure, were related to each other in structurally complex mixed forests. We also asked how well they can predict the light-demand signature of the forest understorey (estimated as the mean Ellenberg indicator value for light [“EIV_LIGHT”] and the proportion of “forest specialists” [“%FS”] within the plots). Furthermore, we asked whether accounting for the shade-casting ability of individual canopy species could improve predictions of EIV_LIGHT and %FS. Location: A total of 192 study plots from nineteen temperate forest regions across Europe. Methods: In each plot, we measured stand basal area (all stems >7.5 cm diameter), canopy closure (with a densiometer) and visually estimated the percentage cover of all plant species in the herb (<1 m), shrub (1–7 m) and tree layer (>7 m). We used linear mixed-effect models to assess the relationships between basal area, canopy cover and canopy closure. We performed model comparisons, based on R² and the Akaike Information Criterion (AIC), to assess which stand characteristics can predict EIV_LIGHT and %FS best, and to assess whether canopy shade-casting ability can significantly improve model fit. Results: Canopy closure and cover were weakly related to each other, but showed no relation with basal area. For both EIV_LIGHT and %FS, canopy cover was the best predictor. Including the share of high-shade-casting species in both the basal-area and cover models improved the model fit for EIV_LIGHT, but not for %FS. Conclusions: The typically expected relationships between basal area, canopy cover and canopy closure were weak or even absent in structurally complex mixed forests. In these forests, easy-to-measure structural canopy characteristics were poor predictors of the understorey light-demand signature, but accounting for compositional characteristics could improve predictions.