This talk explores different ways of ``spelling'' a word in a speech recognizer's lexicon and how to obtain those spellings. In particular, we compare using as the source of sub-words units for which to build acoustic models (1) a coarse phonemic representation, (2) a single, fine phonetic realization, and (3) multiple phonetic realizations with associated likelihoods. We describe how we obtain these different pronunciations from text-to-speech systems and from procedures that build statistical decision trees trained on phonetically labeled corpora. We evaluate these methods applied to speech recognition on the North American Business News (NAB) tasks. For this task (with 60K vocabulary and 34M 1-5 grams), we obtain a 21% relative reduction in word error rate by using multiple phonetic pronuncations over single phonemic pronuncations. We also describe recent work on the efficient repesentation of multiple pronunciations for speech recognition as finite-state transducers.