Eight inferences with quantified sentences were judged as valid or invalid. Each inference involved an active and its correlative passive sentence, e.g. ‘Some medicine cures every disease. Therefore: Every disease is cured by some medicine.’ There were four logically distinct pairs of sentences, and each pair was presented twice: once as an inference from active to passive, and once as the converse inference from passive to active. As predicted, what was crucial in evaluating these inferences was not voice but the position of ‘some’. An inference from a premise with ‘some’ in the grammatical subject to a conclusion with ‘some’ in the grammatical object tended to be judged as valid, whereas an inference in the converse direction tended to be judged as invalid. The results of an earlier study of the ambiguity of the sentences provided a good estimate of performance in the present inferential task.