We examined predictions about the vegetation composition of two experimental wetland cells when they were drawn down in 1983. Two approaches were used to make these predictions: extrapolations from vegetation maps and the van der Valk (1981) model of wetland vegetation dynamics. Both sets of data used to make these predictions were collected in 1980 before the emergent vegetation in the cells was destroyed by raising the water level to 1 m above normal. By comparing digitized vegetation maps from 1983 with those from 1980, we determined which vegetation types were present during the drawdown in areas dominated by different preflooding vegetation types. We predicted that each preflooding vegetation type would b.ave only one corresponding drawdown vegetation type. Our prediction was wrong for three of the most widespread preflooding vegetation types. Areas dominated by each of these types in L980 developed two different vegetation types during the drawdown. When both cells were conlidered together, the van der Valk (1981) model, which used the 1980 seed bank data as its primary input, predicted successfully all species that would be present during the 1983 drawdown with a density of one or more seedlings per 10 m2 • However, within an elevation range or within lreas dominated in 1980 by a particular vegetation type, qualitative predictions were less relilble. Even when both cells were considered together, quantitative predictions of average leedling densities were generally too high for most emergent and wet meadow species and too .ow for most annuals. Within an elevation zone or preflooding vegetation type, discrepancies )etween predicted densities and actual densities were even larger. These discrepancies seemed to )e due primarily to differences between environmental conditions in the field in 1983 and in the lhelter where the seed bank study was done in 1980.