We construct symmetric decompositions of bivariate distributions with given mean values into convex combinations of distributions with supports containing not more than three points and with the same mean values. We use these decompositions for constructing representations of bivariate random...