Despite the common use of mean length of utterance (MLU) as a diagnostic measure, what it actually reflects in terms of linguistic knowledge is relatively unclear. This study explored the extent to which variance in MLU could be accounted for by a measure of expressive vocabulary and a measure of morphosyntax in a group of 44 typically-developing children, ages 28-37 months. Regression techniques were employed to predict MLU from number of different words (NDW) and a tense accuracy composite (TAC). Depending on whether the variable was acting alone or in concert with the other, TAC accounted for 23% or none of the variance in MLU, while NDW accounted for 51% or 28%. We offer three potential explanations for the observed association between MLU and NDW, none of which necessarily precludes the others.