This study presents an approach based on the Bayesian paradigm to identify and com- pare observed changes in the timing of phenological events in plants. Previous studies have been based mostly on linear trend analyses. Our comprehensive phenological dataset consists of long-term observational records (>30 yr) within the 1951-1999 period across central Europe, from which we selected 2600 quality-checked records of 90 phenophases (mostly in spring and summer). We esti- mated the model probabilities and rates of change (trends) of 3 competing models: (1) constant (mean onset date), (2) linear (constant trend over time) and (3) change point (time-varying change). The change point model involves the selection of 2 linear segments which match at a particular time. The matching point is estimated by an examination of all possible breaks weighted by their respective change point probability. Generally we found more pronounced changes in maritime Western and Central Europe. The functional behaviour of all 2600 time series was best represented by the change point model (62%), followed by the linear model (24%); the constant model was the least preferred alternative. Therefore, non-linear phenological changes were by far the most commonly observed feature, especially in Western Europe. Regression analyses of change point model probabilities against geographic coordinates and altitude resulted in some significant negative regression coeffi- cients with longitude; in contrast, the constant model probabilities increased with longitude. Even when differences between locations across Europe existed, an overall trend towards earlier flowering was determined at most locations. Multiple regressions confirmed that mean advancing trends in the 1990s were stronger in the northwestern part of the study area.