In some cases the fitness value of a knowledge base is not completely determined, but just bounded in an interval. In this case the fitness value is modelled by a random variable. Thus the comparison of random variables allows to compare the fitness values when they are not completely determined. In this contribution we consider a quite new proposal in stochastic comparison: statistical preference. We recall the advantages of this method with respect to (the classical) stochastic dominance. We also order by statistical preference two fitness values modelled by beta distributions with some special parameters.