Abstract Many children with autism who speak tend to emit utterances of only 1 or 2 words. A multiple baseline design with changing criterion features was used to evaluate the effectiveness of training with pictorial prompts to increase the syntactical complexity and length of novel utterances by 4 preschoolers with autism. Two target sentence structures were taught; responses during probe trials matched the specific sentence structure being trained. At baseline utterances averaged 1–2 words per photograph. All children averaged 6 words per photograph following training. Variability in the number of training trials to mastery across participants was likely due to differences in entry skills. The training methods were effective for establishing one type of generative language performance in the 4 participants.