The present research extends recent work on the prediction of preschool language skills by exploring prediction from decontextualized vocabulary comprehension. Vocabulary comprehension was a stronger predictor than parent-reported production, yielding a quadrupling of variance accounted for relative to prior studies. Parallel studies (Studies 1 and 2) are reported for two linguistically and geographically distinct samples. In both samples, decontextualized vocabulary comprehension late in the second year provided the best balance between model fit and parsimony in predicting language skills at age three. In Study 3, vocabulary comprehension prospectively identified children with low language status 2 years earlier than other prospective studies but with similar sensitivity and specificity. The present paper provides evidence on three questions of practical and theoretical significance: the relation between decontextualized vocabulary prior to 30 months of age and language outcomes, how prediction from decontextualized vocabulary compares with parent-reported vocabulary, and finally how early stable predictions to language outcomes can be made.