Recent studies have found that the relative importance of predictors of metacommunity structure is dependent on different factors. Low explanatory power of multivariate models is a frequent result. To increase this power, ecologists have suggested different strategies, including the use of functional approaches. Using a phytoplankton dataset from 17 reservoirs in Southern Brazil, sampled seasonally over eight years, we tested the hypothesis that the explanatory power of multivariate models would be higher when the analyses were based on functional groups than when based on a taxonomic approach. We also modeled the temporal variation in the strength of species sorting (as given by the adjusted coefficient of determination derived from environmental variables). We found high temporal variability in the strength of species sorting, indicating that results from snapshot surveys should be interpreted cautiously. When compared to the taxonomic approach, we did not find an increase in the explanatory power of multivariate models when the analyses were based on a functional approach. The main correlates of the temporal variation in the strength of species sorting were insolation, water temperature, and environmental heterogeneity, suggesting that conditions related to productivity and heterogeneity are important in determining the role of species sorting in phytoplankton communities.