Species composition and turnover that have occurred in a series of permanentsample plots established during the 1930s and 1940s in Budongo, asemi-deciduous Ugandan forest, are reported. The plots were establishedas part of a sequence first used to describe forest succession, five ofwhich have been maintained and which were last measured in 1992-1993. Oneplot (plot 7) provides 53 y of data from old-growth pristine forest. Plot15 was established in wooded grassland at the forest edge and is now closedhigh forest. Evaluation of the remaining three plots is complicated bysilvicultural interventions carried out in the 1950s. Forty species havebeen added since the first evaluations and a total of 188 tree species (over80% of Budongo's forest tree flora, and including two exotics) has nowbeen recorded from within the plots. The pattern of shade-tolerance in theoriginal plot series conforms to patterns expected for succession with anincreasing proportion of shade-tolerant species with development, and largestems appearing to ‘lag behind’ smaller stems in this respect. The timeseries data are less consistent, and while plot 7 increased in theproportion of shade-tolerant stems through time, the proportion ofshade-tolerant species actually declines. Stem-turnover (the mean ofmortality and recruitment) slowed with implied successional stage. Mostspecies have a higher recruitment than mortality rate and stem numbers havethus increased in all plots. This is most pronounced in the putatively‘early successional’ plot. Stem size structure has changedwithin the plots, with an increased proportion of smaller stems. Speciesshow different rates of turnover and these vary from plot to plot and periodto period. In plot 7, the overall mortality rate decreased with initial stemsize. Estimates imply that some tree species may easily live longer than 500y after reaching 10 cm DBH, and that 1000 y is possible. The importance oflarge trees in determining forest dynamics is illustrated by the findingthat death of only seven stems in plot 7 contributed over 60% of net basalarea losses recorded over the 53-y observation period. Many of the observedpatterns were not predicted and could only have been found by long-termstudies.