It was hypothesized that prenatal exposure to cocaine and other substances would be related to delayed expressive language development. Speech and language data were available for 458 6-year olds (204 were exposed to cocaine). No significant univariate or multivariate differences by cocaine exposure group were observed. Classification and regression tree modeling was then used to identify language variable composites predictive of cocaine exposure status. Meaningful cut points for two language measures were identified and validated. Children with a type token ratio of less than 0.42 and with fewer than 97 word types were classified into a low language group. Low language children (n = 57) were more likely to be cocaine exposed (63.1%), with cocaine-exposed children 2.4 times more likely to be in the low language group compared with control children after adjustment for covariates. Prenatal cigarette, but not alcohol exposure, was also significantly related to expressive language delays.