In this paper, an unknown word processing method is investigated as an approach to achieve the robust spontaneous speech understanding. The unknown words are detected by a subword-unit based decoder and the process is incorporated into the One-Pass search algorithm. The preliminary experiments showed that the method is also effective to the utterance including interjections (or filled pauses). In addition, the phrase-spotting based approaches were compared with the One-Pass search method in which the unknown word processing was incorporated. The experiments showed that the One-Pass method attained the best performance on spontaneous speech.