The invention proposes a novel information-entropy-based vertical mill running step fault feature extraction and diagnosis method. The method comprises: sample data of a normal working condition and afault working condition during running of a vertical mill are extracted; an information entropy of sample data is calculated; an entropy difference value of the normal working condition and the faultworking condition is calculated; a Fisher classifier is established; a training sample and a test sample are inputted into the Fisher classifier for classification; the normal working condition and the fault working condition are clustered; and the test sample is inputted into the Fisher classifier for diagnosis. Therefore, the feature of the running feature of the vertical mill can be extractedeffectively and accurately; and the diagnosis result has relatively high precision.