Coarticulation is one of the major factors that make speech recognition difficult. In conventional connected word recognition methods, a template for a connected word sequence is made by simply concatenating the templates of a single word. Therefore, some misrecognitions occur on account of this disregard for coarticulation. This study examines the influence of coarticulation in concatenating templates. The boundaries between words are smoothed so as to incorporate the influence of the tail of the preceding word on the head of the following word. The results of recognition experiments of 35 tokens of 4-digit-sequences