One of the most important problems in continuous speech recognition is to accurately model the acoustic variabilities caused by coarticulation in order to obtain high recognition accuracy. In this paper, a novel approach to solve the problem for compensating inter-syllable coarticulation effect on syllable-based continuous Mandarin speech recognition is proposed. Inter-syllable coarticulation is directly modelled by supplementing the context-independent syllable HMM models with some inter-syllable boundary models. Experimental results showed that the syllable recognition errors of a speaker-dependent continuous speech recognition task was reduced by 24%.