Since Upshur's (1969) original paper describing a picture-description test of oral communication ability, four empirical studies have been completed in which variants of the tests have been used. From these studies considerable new information on the tests' reliability and validity has become available. Indications are that the reliability is somewhat less than originally estimated and that concurrent validity with the oral interview is quite low. A feature analysis of the speech behavior required by the picture-description tests indicates a number of abnormalities that could account for the tests' low validity. The model that was developed to analyze the tests sees production as varying from pure manipulation at one extreme, through noncommunicative (yet meaningful) use, to autonomous communication at the other extreme. The features involved are: propositional content, speech-act content, information sequence, uncertainty, intent, processing, and shared reference. The implication drawn from the analyses is that if controlled tests of communication are needed, an effort should be made to minimize the effect of controls on the naturalness of the speech behavior. (Author/AMH)