The invention discloses a user input identification method based on a mobile phone inertia sensor and realizes the purpose of collecting, training, detecting and identifying data under the condition that a user is unaware. The method comprises the steps of generating attitude angle information by fusing data of an accelerometer and a gyroscope; extracting classification features for identifying the user input from the attitude angle information; training a plurality of different types of classifiers; establishing a combined classifier through adoption of a combination voting strategy; in an identification phase, classifying attitude features through utilization of the combined classifier; identifying single characters; establishing a hidden Markov model; and eliminating potential password combinations not satisfying an observing sequence by taking a sampling interval of the accelerometer as the observing sequence. The technical problem that an account password cannot be identified through application of the existing user input identification method is solved.