Our proposed work is aimed at teaching non-native Arabic speakershow to improve their pronunciation. This paper reports on a diagnostic tool forhelping non-native speakers of Arabic improve their pronunciation, particularlyof words involving sounds that are not distinguished in their native languages.The tool involves the implementation of several substantial pieces of software.The first task is to ensure the system we are building can distinguish betweenthe more challenging sounds when they are produced by a native speaker, sincewithout that, it will not be possible to classify learners’ attempts at these sounds.To this end, we carried out a number of experiments with the well-known speechrecognition Hidden Markov Model Toolkit (HTK), in order to ensure that itcan distinguish between confusable sounds, such as the ones that people havedifficulty with. Our diagnostic tool provides feedback in three different forms: asan animation of the vocal tract, as a synthesised version of the target utterance,and as a set of written instructions. We have evaluated the tool by placing it in aclassroom setting, asking 40 Arabic students to use the different versions of thetool. Each student had a thirty minute session with the tool, working their waythrough a set of pronunciation exercises at their own pace. Preliminary resultsfrom this pilot group show that their pronunciation does improve over the courseof the session.