Abstract Rolling element bearings are one of the fundamental components of a machine, and their failure is the most frequent cause of machine breakdown. Monitoring the bearing condition is vital to preventing unexpected shutdowns and improving their maintenance planning. Specifically, the bearing vibration can be measured and analyzed to diagnose bearing faults. Accurate fault diagnosis can be achieved by analyzing the envelope spectrum of a narrowband filtered vibration signal. The optimal narrow-band is centered at the resonance frequency of the bearing. However, how to determine the optimal narrow-band is a challenge. Several methods aim to identify the optimal narrow-band, but they are not always precise. The bearing fault vibration components are lost if the narrow-band is incorrectly chosen, thus leading to an incorrect fault diagnosis. For on-line systems, it is critical that bearing faults are diagnosed with a high degree of confidence. In this article, a method for analyzing multiple narrow bands is presented. Bearing faults are detected autonomously by a narrow-band envelope spectrum-based algorithm. This algorithm removes the need for manual spectrum analysis, allowing operators to focus on more important tasks. Bearing fault vibration data from an accelerated life-test is used to verify the performance of the proposed method. The proposed method accurately diagnoses the worn-out bearing for three characteristic defect types and shows when one fault propagates to a second one.