This paper describes a system for the inductive learning of grapheme-to-phoneme rules. As input, the system only needs a list of words; each word in its orthographic and phonemic form. In a first step, the correspondence between graphemes and phonemes is established for each word. This is done with the technique of Hidden Markov Models. Next, the actual learning process is started. This iterative induction process creates an ordered list of pronunciation rules for each letter of the alphabet. The proposed methods were evaluated for Dutch.