Due to its characteristics of shortness and sparseness, short text, as the main body of microblog and other social media, cannot be accurately classified by the traditional text classification methods. To solve this problem, a method of short text classification based on association rules of lexical items was proposed in this paper. Firstly, the training set based on the strong association rules was mined, and then the strong association rules was added to the features of short text so as to increase the feature density of short text, thereby to increase the accuracy of results of short text classification. Comparative experiments show that this method, to some extent, reduces the impact of sparseness of short text on the classification results, and it improves the classification accuracy, recall values and F1 values.