Individual differences in generating causal backward inferences during texts comprehension may reflect differences in reading comprehension. The methods to investigate causal inferences, used in this research are indirect (the cued recall and asking-answer protocol), because of the phenomenological impenetrability on-line inference generation. Exp 1 shows differences between concret operatory stage and formal operatory stage subjects, in the efficacy of retrieving consequent pairs of sentences with a variable strengh of the causal relationship. Exp 2 is a longitudinal study which presents three types of causal backward inferences, generated when antecedent-consequent relationships are congruent /incongruent with the contextual information. Our results shows that concret operatory stage subjects have lower performances for generation reactivated inferences and elaborations, than formal operator stage subjects.