The invention discloses a method for detecting abnormal messages. The method comprises the steps that a text with the messages to be detected is divided into a plurality of text segments; one or more account number attributes of each text segment are obtained and publishing ratio parameters corresponding to each account number attribute of each text segment are determined; according to the publishing ratio parameters, a first factor, corresponding to each account number attribute, of each text segment is determined; according to the first factor, corresponding to each account number attribute, of each text segment, a second factor is determined; according to the second factors, whether the detected messages are abnormal messages or not is determined. The publishing account number attributes and indiscrimination text segmentation processing are combined and the Bayesian algorithm is used, so that effective limit on microblog spam in batches is achieved, and meanwhile the flexibility of spam processing is improved.