Air-traffic has grown rapidly in the last twenty years and concern has been mounting about the safety implications of mis-recognition of call-signs by both pilots and air-traffic controllers. This paper presents the results of a preliminary study into perceptual (i.e. non-cognitive) confusions in two closed vocabularies of the type used as aircraft call-signs. Conventional methods of subjective and objective testing were found to be unsuitable for our aim of predicting potential confusions within a vocabulary. Hence a method for modelling confusion probability in a closed vocabulary at a certain signal-to-noise ratio has been developed. The method is based on the use of a phoneme confusion matrix and a technique for comparing phoneme strings. The method is presented and results are given. These suggest that the behaviour of the model is plausible, and a comparison of its predictions with a set of real confusions showed a correct prediction of position of confusion in three-word phrases. The predictions of the model need to be verified by subjective testing before it can be deployed in a system that designs low-confusability call-signs, which is the ultimate goal of the research.