This paper introduces a novel frequency pattern for diagnosis of broken rotor bars in induction motors during the start-up period. This is important in the case of start-up of large induction motors with long starting time and also motors with frequent start-up. The amplitude of the side-band components of the introduced frequency pattern is a suitable index for diagnosing the broken bars and determining the number of faulty bars precisely during start-up. The paper investigates the performance of a faulty motor. Because the number of broken rotor bars and the load are two major parameters affecting the performance of a faulty motor, the frequency spectrum of torque and speed in a motor with different numbers of broken rotor bars and different loads is analyzed. The time-stepping finite-element method (TSFEM) is used to model the motor. This model takes into account the spatial distribution of the stator windings, nonuniform magnetic permeance of the air gap, nonlinear characteristic of the cores, and continuous rotation of the rotor. The paper compares simulation results in steady-state mode with experimental results and shows that the results agree well.