Aims: Inconsistent treatment of the vegetation layers is one of the main problems in the floristic classification of forests. In this study I investigate whether a classification based solely on woody species leads to units similar to the Braun-Blanquet system or to something completely different. Study area: Austria (Central Europe) and adjacent regions. Methods: 23,681 forest relevés from the Austrian Vegetation Database were classified using TWINSPAN. Spruce and pine plantations and stands with a cover of non-native woody species &gt; 5% were excluded from the dataset. Only native tree and shrub species were used in the classification while herbs, dwarf shrubs, cryptogams and all records of woody species in the herb layer were omitted. Results: The TWINSPAN classification revealed elevation (i.e., climate) as the main floristic gradient in the data set. Within lowland communities, soil moisture was the dominant factor. The higher units of the Braun-Blanquet system were mostly well reproduced. Conclusions: The higher levels of the phytosociological forest classification (class, order, partly also alliance) can basically be defined by taking only the shrub and tree layer into account. However, all past and current classifications suffer from arbitrary exceptions to this rule. This leads to many inconsistencies and blurs the main biogeographical patterns within European forests. Here I argue that using the tree and shrub species for defining the higher levels and the understorey species for defining the lower ones is best suited to meet the properties that users would expect from a good forest classification.<br> Taxonomic reference: Fischer et al. (2008).<br> Syntaxonomic reference: Mucina et al. (2016) if not stated otherwise.<br> Abbreviations: EVC = EuroVegChecklist (Mucina et al. 2016).
