Differences in the early development of children are crucial for early detection of autism spectrum disorder (ASD). Previous studies have shown large differences between children later diagnosed with ASD and their typically developing peers in the early use of canonical vocalizations (i.e., vocalizations that include well‐formed consonant–vowel syllables) and the use of vocalizations for communicative purposes. In this prospective study, we examined the extent to which infant vocalizations at 14 months would predict Autism Diagnostic Observation Schedule (ADOS) diagnostic symptom groups, that is, Autism, Spectrum, and Non‐ASD, for 82 community‐identified at‐risk infants at 23 months. Thirty‐minute video samples were coded with the intention to categorize and quantify speech (canonical/noncanonical and directed/nondirected) and nonspeech vocalizations (atypical, distress, and pleasure vocalizations). Our results revealed that more canonical directed (OR = 1.039, P = .036), and fewer noncanonical directed (OR=.607, P = .002) and noncanonical nondirected (OR = 1.200, P = .049) vocalizations were associated with a greater likelihood of being in the Non‐ASD group versus the Autism group, with no variables significantly predicting Autism versus Spectrum group membership. Despite some statistically significant findings, models performed poorly in classifying children into correct ASD symptom group at age 23 months based on vocalizations at 14 months. Thus, the utility of infant vocalizations alone for predicting toddler clinical outcomes among infants initially identified at an elevated risk for ASD appears limited; however, considering the structure and function of early vocalizations combined with other early developmental and behavioral features may improve the confidence for clinicians in making an early diagnosis of ASD. Autism Res 2017. © 2017 International Society for Autism Research, Wiley Periodicals, Inc.