For most experimental studies the short-term responses to manipulation often differ from the long-term changes in the community composition, dynamics or functioning. Such discrepancy limits the translation of experimental results into key ecological topics such as global environmental change. Here we analyzed plant community dynamics from a 23-year transplant experiment in the Fennoscandian mountain tundra and explored how well the pattern of responses over the first 12 years of the experiment can predict longer-term changes. Sod-blocks of tundra heath vegetation were transplanted to a snowbed 150 m higher in elevation from their origin, where, with contrasting levels of soil wetness, half of the transplants were protected from mammalian herbivores. Throughout the experiment, community changes strongly depended on both plant functional types and experimental treatments. The first 12 years were characterized by a response to transplantation to the snowbed showing a strong increase of graminoid and a decrease of shrub abundances in the transplants. In the longer term, the community divergence increased in particular in response to grazing and soil wetness within the snowbed, while graminoid dominance disappeared. Markov chain models captured the main trends during the first 12 years but they failed to predict their relative abundance after 23 years. In particular, the late dominance of bryophytes in the wet snowbed, the recovery of shrubs in the dry exclosures, and the subordinate status of graminoids deviated from the extrapolation based on the medium-term trends. Despite clear community dynamical trajectories detected in the first decade, the differences in the temporal scale of both treatment effects and plant functional type responses limited their ability to extrapolate longer-term trajectories. We find that increasing focus on long-term experiments is a crucial step to understanding the processes involved in the response of plant communities to global environmental change.