Preschoolers' earliest productions of sentential complement sentences have matrix clauses that are limited in form. Diessel proposed that matrix clauses in these early productions are propositionally empty fixed phrases that lack semantic and syntactic integration with the clausal complement. By 4 years of age, however, preschoolers produce sentential complement sentences with matrix clauses that are more varied. Diessel proposed that the matrix clauses in these later productions semantically and syntactically embed the complement clause. We refer to these matrix clauses as formulaic and true, respectively. Diessel's hypothesis about the development of sentential complement sentences was based on an analysis of spontaneous language. The purpose of this study was to evaluate Diessel's hypothesis with an experimental sentence imitation task wherein stimuli varied in the nature of the matrix clause. Thirty children with typical language development participated; 10 children in each age group (3-, 4-, and 5-year-olds) imitated 50 sentential complement sentences that included either a true or a formulaic matrix clause; the structure of the dependent clauses did not vary. Dependent variables were percent sentence imitation and percent matrix clause imitation. There was a significant main effect for matrix clause type on imitation of sentences and matrix clauses. There was also a significant main effect for age on imitation of sentences and matrix clauses. Significant matrix clause type-by-age interactions were such that percent sentence imitation and percent matrix clause imitation varied by age. Three- and 4-year-olds were less proficient than 5-year-olds on imitation of sentences with true matrix clauses and on imitations of true matrix clauses. Only 3- and 4-year-olds were less proficient imitating true matrix clauses than formulaic matrix clauses. Experimental findings support Diessel's hypothesis that there is a developmental progression in the nature of preschoolers' production of sentential complement sentences.