After a critical examination of four assumptions of current procedures in language assessment, it is argued that the pragmatic domain of language ability should be the first to be assessed. A study designed to determine the usefulness of language variation (the number of different structures a child can produce or comprehend) as an alternative approach to language assessment is discussed. Thirty-two elementary school children with Hispanic backgrounds were classified as balanced or non-balanced bilinguals on the basis of performance in the Peabody Picture Vocabulary Test. Request strategies in English and Spanish were elicited with the use of puppets in simulated settings. Balanced bilinguals produced more variation in their second language than non-balanced bilinguals. This finding supports Cummins' 1978 linguistic interdependence hypothesis by showing that a relationship exists between a bilingual speaker's two languages and that this relationship influences performance on certain tasks in the second language. Implications for bilingual education are briefly discussed, and a bibliography is appended. (JB)