Fault detection in alternating-current electrical machines that is based on frequency analysis of stator current has been the interest of many researchers. Several frequency estimation techniques have been developed and are used to help the induction machine fault detection and diagnosis. This paper presents a technique to improve the fault detection technique by using the classical multiple signal classification (MUSIC) method. This method is a powerful tool that extracts meaningful frequencies from the signal, and it has been widely used in different areas, which include electrical machines. In the proposed application, the fault sensitive frequencies have to be found in the stator current signature. They are numerous in a given frequency range, and they are affected by the signal-to-noise ratio. Then, the MUSIC method takes a long computation time to find many frequencies by increasing the dimension of the autocorrelation matrix. To solve this problem, an algorithm that is based on zooming in a specific frequency range is proposed with MUSIC in order to improve the performances of frequency extraction. Moreover, the method is integrated as a part of MUSIC to estimate the frequency signal dimension order based on classification of autocorrelation matrix eigenvalues. The proposed algorithm has been applied to detect a rotor broken bar fault in a three-phase squirrel-cage induction machine under different loads and in steady-state condition.