Japanese has complex verb formations that convert two verbs into a complex predicate. A salient type among them is the compound verb that is made of two verb stems, V1 (renyooform=continuative form) and V2 such as nomiaruku (drinking walk=barhopping). Although the compound verbs are productive it is often diffi cult for non-native learners to predict their meanings because there are various ways of semantic combinations between V1 and V2. And even worse for Chinese students learning Japanese, Chinese compound verbs that correspond lexically to Japanese compound verbs are sometimes different in meaning, which may cause errors in acquisition. As a result, learners tend to avoid them and instead to use te-form verbs composed by te-form V1 and V2 because they are easy to build. But this strategy is not successful when te-form verbs don't mean the same meanings as the counterpart compound verbs. In this paper we try to find out the factors of error usage of Japanese compound verbs by learners based on error analysis and to classify their semantic properties (syntactic and lexical compound verbs and classifi cation by LCS linking etc.). Furthermore, we compare compound verbs and te-form verbs with respect to the semantic hierarchy and show at which level they are similar and at which level they differ. Specifi cally, lexical compound verbs cannot be paraphrased by te-forms when V1 or V2 has a metaphorical or formalized meaning.