In this thesis, we studied the corpus and classified the types of errors in the spoken discourse, and use the statistical to make tables. Compared and made analysis the college students of different grades, we tried to find the differences of the error types in the spoken discourse, described them and analyzed its evolution trend. From the research, we found that the means of connection and omit are most commonly used in spoken discourse, but the mean of words the students did not used much, in addition, from the statistics, we found the errors on coherence are mainly caused by using the same words over repeated. At last, through the evolution of trend, we found that as the Chinese levels higher and higher the errors which the students made are fewer and fewer, except using the mean of omit. The conclusions are very important for making teaching program and defining the teaching emphasis. This thesis also provide some useful suggestions and comments on teaching methods and teaching materials editing.