An extension of Luce's choice analysis of recognition experiments has been made so as to predict quantitatively the syntactic confusions that occur among simple syntactically related sentences which are heard at several times normal conversation rate (speeded speech). Two states are assumed to account for the pattern of asymmetric confusions: (a) a dimensional loss state wherein dimensional cues of the stimulus sentence are lost on a certain proportion of trials, and (b) a nonloss state wherein all relevant sytactic cues are assumed to be sampled but where confusions among stimuli are still assumed to occur in a predictable manner. Some relationships between the present model and Chomsky's distinction between competence and performance are discussed.