This paper presents a feature selection method using mutual information to identify the most relevant features, applied for fault diagnosis of induction motors. Time-domain features extracted from vibration data of induction motors are used as input data for fault diagnosis process. The dataset, formed by the most significant features selected by the proposed method, is divided into training and testing sets and evaluated using supervised learning techniques such as k-nearest neighbors and support vector machine. Performance evaluation on the test data demonstrates the potential and effectiveness of this selection method for diagnosing faults in induction motors.