An objective method for deciding which effects are active in the analysis of unreplicated experiments, when the model is saturated and hence there are no degrees of freedom for estimating the error variance.