A semantic Web service discovery based on WordNet ontology and PLSA (Probabilistic Latent Semantic Analysis) was proposed. Firstly, the operation name, input parameters, output parameters and the query request were annotated by WordNet ontology. Then the annotated output parameters were taken as term set, and web services as document set to construct a vocabulary-document matrix. After that the query request was taken as service and was injected into the PLSA model to determine its category. Secondly, in this category, a mapping table was established. The output parameters of services were taken as map key, and services list which contained the output parameter was taken as the key value for the map. Each output parameter of query request could match the map keys which were similar to themselves. According to the matched map key, the corresponding list of services would be identified, and then the service set that was compatible with QoS requirements also would be identified. Finally, the final service result set would be acquired according to the query request input parameters. Testing data set including 415 web services was used to make an experiment for this method. Results show that not only the performance is better than other methods but also ��R��-precision and recall rate have been improved.