Long-term time series are increasingly used to assess the effects of global change on plant community diversity and to guide management of target plant communities. However, historical biodiversity data may contain neglected sources of error that can have a significant impact on the results and their interpretation. In our study, we focus on historical sampling error, a source of potential bias in long-term biodiversity assessments that has not been systematically addressed. We resampled two historical datasets of a different origin in the floodplain forests of the Czech Republic, with 534 vegetation plots originally sampled in the 1950s and 1960s. We compared temporal trends in alpha diversity and Ellenberg indicator values (EIVs) between the two parallel surveys. To assess compositional differences, we compared temporal changes in species frequencies. Alpha diversity increased by 9.3 % in one resurvey, but decreased by an average of 30.8 % in the second resurvey. The distribution of EIVs for plots also differed, indicating that each resurvey covered a different part of the environmental gradient. We conclude that preferential historical sampling of the vegetation-environment continuum and species omission may have contributed to the differences in biodiversity and environmental change between the datasets. Our study shows that historical sampling error can have a significant impact on assessments of long-term biodiversity trends. We recommend that historical reference datasets should be critically assessed for potential sources of error in assessments of environmental change and management objectives.