This paper describes a Chinese spoken language understanding system USTC-2. Various kinds of knowledge such as acoustic-phonetics, vocabulary, syntax and semantics are represented and utilized in the system. It's constructed as an expert system based on frame representation capable of metting the needs of various tasks. In USTC-2 system, segmentation of input speech is carried out utilizing the energy contour, LPC variance and syllable length. Recognition is performed by a matching algorithm which constrains the search on the basis of morphological knowledge concerning both the part of word-class and and constituent syllales of a given word . The use of morpohological knowledge not only reduces the amount of computation but also increases the correct rate of syllable recognition in spoken Chinese sentences. An analysis method starting from KEYWORDS was proposed in speech understanding. This method combines syntastic analysis and semantic analysis together and has strong ability of correcting errors. The system discriminates homonymous syllables and homonymous words in each processing step.