A method for predicting prison term based on multi-task artificial neural network includes such steps as analyzing and mining the data of judgment document, predicting the term of imprisonment of newly obtained description of crime fact and basic information of suspect, predicting the term of imprisonment of newly obtained description of crime fact and basic information of suspect, predicting theterm of imprisonment of newly obtained description of crime fact, predicting the term of imprisonment of newly acquired description of crime, predicting the term of imprisonment of newly obtained description of crime and basic information of suspect, predicting the term of imprisonment of newly obtained description of crime, and predicting the term of imprisonment of newly obtained suspect. Guidedby the attributes of legal clauses cited in judgment documents, this method constructs a multi-task neural network model with mutual support effect by making full use of the multi-dimensional data injudgment documents, and trains the pre-processed judgment documents data to obtain a predictive method with high precision and strong practicability. This method takes 21 legal articles such as accusation, degree of injury and amount of money involved as auxiliary tasks, and classifies life sentence and death sentence separately, so as to achieve accurate prediction of sentence. Compared with thesingle-task neural network model based on such attributes, the prediction accuracy rate of the method proposed by the invention is higher, indicating that the method proposed by the invention is effective and practical.