The passive construction is a late acquisition in child language. In this paper I evaluate the claim that difficulty with noncanonical semantics, rather than non-mature subject-object A-chains, underlie young children’s poor performance on the passive. In a series of truth-value judgment tasks, 4- and 5-year-old English-speaking children were tested on their comprehension of matrix passives and passives embedded under raising-to-object (RO: want, need) and object control (OC: ask, tell) verbs. RO passives (Olivia wanted/needed Scott to be kissed by Misha) entail object-subject-object A-chains, but allow semantic patients to surface as syntactic objects; OC passives (Olivia asked/told Scott to be kissed by Misha) and matrix passives (Scott was kissed by Misha) involve similar A-chains but do not result in this syntactic-semantic configuration. I found that although 4-year-olds failed to comprehend matrix passives and passives embedded under OC verbs, they correctly interpreted passives under RO verbs. (5-year-olds performed above chance in all tasks.) I propose a “semantic scaffolding” account of children’s comprehension of the passive that rests on the prototypicality of subjects being agents and objects being patients, arguing against the view that difficulty with passives results from an inability to form A-chains.