Objective: This article examines the form and function of spontaneous communication and outcome predictors in nonverbal children with autism following classroom-based intervention (Picture Exchange Communication System [PECS] training). Method: 84 children from 15 schools participated in a randomized controlled trial (RCT) of PECS (P. Howlin, R. K. Gordon, G. Pasco, A. Wade, & T. Charman, 2007). They were aged 4-10 years (73 boys). Primary outcome measure was naturalistic observation of communication in the classroom. Multilevel Poisson regression was used to test for intervention effects and outcome predictors. Results: Spontaneous communication using picture cards, speech, or both increased significantly following training (rate ratio [RR] =1.90, 95% CI [1.46, 2.48], p less than 0.001; RR = 1.77, 95% CI [1.35, 2.32], p less than 0.001; RR = 3.74, 95% CI [2.19, 6.37], p less than 0.001, respectively). Spontaneous communication to request objects significantly increased (RR = 2.17, 95% CI [1.75, 2.68], p less than 0.001), but spontaneous requesting for social purposes did not (RR = 1.34, 95% CI [0.83, 2.18], p = 0.237). Only the effect on spontaneous speech persisted by follow-up (9 months later). Less severe baseline autism symptomatology (lower Autism Diagnosis Observation Schedule [ADOS] score; C. Lord et al., 2000) was associated with greater increase in spontaneous speech (RR = 0.90, 95% CI [0.83, 0.98], p = 0.011) and less severe baseline expressive language impairment (lower ADOS item A1 score), with larger increases in spontaneous use of speech and pictures together (RR = 0.62, 95% CI [0.44, 0.88], p = 0.008). Conclusion: Overall, PECS appeared to enhance children's spontaneous communication for instrumental requesting using pictures, speech, or a combination of both. Some effects of training were moderated by baseline factors. For example, PECS appears to have increased spontaneous speech in children who could talk a little at baseline. (Contains 3 figures and 2 tables.)