In this work, a probabilistic mobile text entry system based on statistical language models is proposed. Mobile devices having limited keyboards usually produce ambiguous inputs due to multiple letter assignment to each key. Faster typing can be achieved by predicting the intended word which decreases the number of required keystrokes. Currently available methodologies mostly rely on dictionaries, which are impractical for agglutinative languages. The complex morphological structure of these languages gives rise to very high number of word forms that cannot be efficiently covered by any dictionary. The proposed system uses n-gram letter probabilities and K Best Viterbi decoding to generate a list of predictions. The dictionary based method and the proposed probabilistic system are evaluated against a typical agglutinative language, Turkish. The experimental results indicate that the proposed system outperforms the dictionary based method with a 33% improvement in performance(1).