Departing from the reduction of compound lotteries axiom on multi-stage lotteries, this paper proposes a new hybrid model to analyze decision trees. Applied to multi-stage decision trees induced by experiments, Blackwell’s (1953) definition of the relation “more informative” on the set of information structures is equivalent to experiments being more valuable to a class of non-expected utility preferences. This result extends Blackwell’s theorem and provides new insights regarding the evaluation of information produced by experiments.