Current practice in language testing has not yet integrated classical test theory with assessment of language skills. In addition, language testing needs to be part of theory development. Lack of sound testing procedures can lead to problems in research design and ultimately, inappropriate theory development. The debate over dimensionality of language and the testing of proficiency illustrates these difficulties. The introduction of confirmatory analysis should improve research on second language learning. In this paper the confirmatory use of a latent trait model (the "partial credit model") is demonstrated as a tool in the development and construct validation of an oral interview test. The model describes the relationship between an individual's proficiency and the difficulty of a language task, allowing for at least two categories of performance, in terms of the probability of a person providing a language sample adequate to earn a given score within a given limit. It was chosen because of its apparent consistency with observations over a wide range of classroom activities. Item analysis and model-to-data fit were conducted on a 29-item interview test given to 270 students. Use of the approach and model was found to be appropriate and valid. A 33-item bibliography is included. (MSE)