Abstract: It is difficult to convert graphemes to phonemes in German Speech Synthesis system.A rule-driven algorithm based on an iterated finite state transducer is proposed in present research to automatically convert graphemes to corresponding phonemes.In this algorithm，grapheme-to-phoneme rules are first made on the basis of a given lexicon；then graphemes of words are converted to phonemes following iterations resulting from morphological rules and phonological rules.Test results show that the accuracy of the total lexicon following application of the graphemes to phonemes conversion algorithm reaches as high as 94.4%.