This paper is a test of two widely held assumptions in the practice of riparian restoration: (1) if physical processes are restored, plant communities will naturally reassemble themselves, and (2) restored communities will resemble reference sites. Seasonal flooding was restored to two interconnected floodplains in the Central Valley of California (USA), and plant community establishment was studied for six years at 300 permanent vegetation plots. If these two assumptions are valid, then the two floodplains should end up with similar plant assemblages, and they should both have followed a similar trajectory. Then, once the relevant physical processes are restored, (1) plots with similar environmental conditions should have increasingly similar species compositions, (2) plant communities should become more stable and cohesive, (3) both species distributions and plant communities should respond to changes in environmental conditions, (4) plot diversity should decrease, and (5) perennial species should replace annuals. The plots were classified into communities using TWINSPAN, and these communities differed significantly with respect to the main environmental gradient (inundation). BrayCurtis similarities were calculated for each pair of plots. Patterns in similarity were used to test the strength of communities and the relative importance of proximity and inundation. On the northern floodplain, there was a trend of increasing similarity for plots with similar environmental conditions over the course of the study; plant communities became more stable and clearly responded to changes in environmental conditions. Plot diversity decreased, and the proportion of perennial species increased. On the southern floodplain, however, plots with similar environmental conditions became less similar, while plots that were close together became more similar; plant communities did not become more stable though they did shift in response to changes in environmental conditions. Taken together, this evidence suggests that assembly of communities is more stochastic than deterministic.