The use of observation of natural language interaction as a measure of language proficiency and the impact of discourse characteristics on children's use of Spanish and English as measured by length of utterances are examined. The goal of this observational approach to measuring language proficiency is to distinguish between the effects of change in discourse contexts and change in the language proficiency of individuals over time. Maximum likelihood techniques are used to estimate the effects of discourse contexts on length of utterance, and the probability that utterances will be as long as those observed in each discourse context is then calculated. This probability becomes the basis for constructing a weighted index of utterance length. This approach was tested on language samples obtained from Spanish/English bilingual children between the ages of 4 and 10 and compared to other indicators of language ability. It was found that discourse function or context appears to influence utterance length. (Author/RW)