Automatic Speech Recognition (ASR) can be applied to children language learning as a diagnostic tool for mispronunciations or speech therapy purposes. This is a well known challenge, not only because of the linguistic variability of children's speech, when compared to adults’, but especially because of its acoustic characteristics. This study proposes an approach to detect mispronunciations or speech disfluencies in young children. The actual phoneme sequence and the likelihood values (provided by the Viterbi decoding of the speech signal) and the correct phone sequence are compared giving rise to an indication whether the word was mispronounced or not. Results show that the likelihood value may be used as a fairly good indicator of speech disfluency and other articulatory difficulties.