Predictions based on the full-profile method of conjoint analysis can be significantly improved by imposing a priori constraints on the estimated parameters. The a priori constraints are based on relative preferences for different levels within attributes. Constrained estimation, however, has little effect on predictions from the trade-off method since the relative preference judgments within rows (or columns) of trade-off tables play essentially the same role as the constraints.