In vegetation science, data from permanent plots are of great importance for understanding long-term processes and their underlying causes. The longer a permanent plot exists, however, the more difficult it becomes to interpret the observed changes. There are no longer any truly comparable replications in space and, consequently, no information is available on the natural variability of the system studied. Time series from individual plots can, nevertheless, be properly and successfully analysed and interpreted, provided an adequate system of reference is available. 
In the present paper, a regionally defined system of reference is proposed. It consists of statistically derived vegetation units, characterised by different probabilities with which the different plant species occur. Data from a individual permanent plot can now be analysed and interpreted by comparing every relevee of the time series with every vegetation unit of the reference system. Vegetation has changed during the period studied, if the probabilities of belonging to the different units of the reference system have changed significantly. 
The proposed approach was tested using data from permanent plots in the Swiss National Park. The units of the regional reference system "Swiss National Park 1917-1996'' were defined based on the relevees made in approximatly 100 permanent plots established and monitored in the Park from 1917 onwards, using multivariate methods. Since every unit is described by more than 20 relevees, their natural variability in space and rime is well documented. 
The findings were compared with the results obtained using both expert systems of the traditional plant sociology and newer but well established multivariate merthods e.g. based on replications in space or on small number of subjectively selected relevees of reference (fuzzy ordination). No matter what method was used, the conclusions on whether and when a change occurred were suprisingly consistent. 
The proposed regionally defined system of reference, however, represents by far the scientifically most sound and convincing solution for dealing with data from individual permanent plots, especially, when combined with a probability model.