We model trait selection by deriving and comparing models for direct trait selection and selection that is mediated by interspecific interactions. The purpose is to model the two selection forces simultaneously, in order to account for potential trait covariance in the estimation procedure when including multiple trait values. In addition, we identify those traits most important for selection forces, and the role of flooding duration. A Bayesian modelling approach is applied and fitted to plant species cover including stochastic variable selection, to test the importance of individual traits on selection forces. We used pinpoint data from wet grasslands of the French Atlantic Coast. Our results show that trait selection forces are driven by direct selection and, to a minor extent, by selection mediated by interspecific interactions. Of the tested traits leaf dry matter content (LDMC) and specific leaf area (SLA), seemed most important for selection forces. We find a significant effect of flooding on selection forces. Parameter covariance analysis revealed flooding to be most strongly correlated with direct selection forces. The method has less critical assumptions regarding the geographical limit of the ecological environment than a traditional filtering approach, and allows weighting traits simultaneously and accounting for their covariance. Inference is easy to obtain from the posterior parameter distributions of the Bayesian estimation. Thus, the Bayesian method presented is very objective. A program architecture is supplied in the R-programming language.