Novel-word pronunciation poses little problem to the human, yet when a computational text-to-speech (TTS) system is presented with a previously-unseen word, the result is often an incorrect pronunciation. We believe that the lack of success of TTS systems in this respect is a consequence of their over-reliance on pronunciation rules. We have previously described (Sullivan and Damper, 1990) a model of English novel-word pronunciation based on analogy with known words which has its origins in psychological investigations of reading aloud. This paper reports on progress in implementing the model and presents some recent results.