Question: As a result of recent and substantial digitization efforts, herbaria are becoming important sources of data for vegetation scientists. Are such data sets appropriate to describe composition gradients and β-diversity? When compared with phytosociological data, what are the differences in terms of composition (co-occurrence) gradients depending on the considered scale?. Location: Páramos (Neo-tropical alpine ecosystems) of Colombia. Methods: We compared vegetation patterns from phytosociological relevés and reconstructed pseudo-communities from herbarium collections in the Colombian high elevation páramo ecosystem using diversity partitioning and Mantel correlations. Results: Species composition differed in the two data sets, which could be explained by taxonomic bias towards charismatic species and overrepresentation of rare species in the herbarium data set, whereas common species were more frequently represented in the phytosociological data set. The two data sets showed a similarly preponderant importance of large-scale differences when we looked at species accumulation across different scales. Small-scale richness contributed more to total richness for the phytosociological data set, while richness at intermediate scales was more important in the herbarium data set. Finally, pairwise β-diversity analyses did not show correlations between data sets, and common species showed similar ecological distribution patterns. Conclusions: We recommend caution to researchers who wish to describe β-diversity patterns in local communities using only herbarium data. However, since the two data sets showed some complementarity in their composition patterns, we suggest that combining data from relevés (or plots) and occurrence data (herbarium records, citizen science, etc.) could be an efficient strategy for describing broader diversity patterns. We discuss the circumstances under which it could be advantageous to work with such combined data sets, in particular in relation to conservation issues.