Automatic speech recognition is an open research area that applied to make a computer have an artificial ear to listen what human speak using their own language. Different speech recognition research is done for foreign language and local language. But a few efforts are done for local language Afan Oromo. This research develops speaker independent continuous automatic speech recognition system for Afan Oromo language using phones based and syllable based recognition by considering the dialects of the language. To develop ASR system for Afan Oromo we used Hidden Markov Model technique by using HTK tool. The speech corpus is collected from 39 males and 24 females of different dialects and have a length of these speeches is about four hours for training and 40 minutes for testing. Both context independent model using monophone and context-dependent model using triphone, tied triphone state and syllable, Afan Oromo ASR system is developed. Backed off bigram language model, phone-based and syllabus based alternative pronunciation dictionary is also developed for building the recognizer. According to the finding of this research, the performance gained for syllable based Afan Oromo speech recognition is highly promising as the frequency of syllables increases. Finally, we get 39.55%, 47.21, 55.35and 43.96% of the correctly recognized word using monophone, tri-phone, tied state triphone and syllable based recognition model respectively