In this paper we discuss firstly a method for skeltal syntactic analysis which uses only surface information.By this method we can successfully abstract the skeltal syntactic structure of a long Japanese sentence. Secondly we discuss its use on building a large corpus with skeltal syntactic structure consisting of about 80 thousand example sentences from two Japanese-English dictionaries. Thirdly we discuss our development of a similar sentence retrieval system over this corpus. The advantage of this system is that (1) Syntactic structure as input search pattern enables the system to dismiss unnecessary example sentences. Thus it raises the precision of the sentence retrieval.(2) Similar sentence retrieval realized on the basis of word encoding according to the thesaurus hierarchy.(3) Index table including syntactic information enables the system to make high speed retrieval.(4) The system has an excellent graphical user interface using the triangle representation of syntactic structure.