Early fault signals of rolling bearings are relatively weak and often submerged in environmental noise. They are difficult to be recognized and extracted. An early weak fault detection method for rolling bearings based on the Chen system was proposed. Firstly, the nonlinear dynamic characteristics of the Chen system were studied, and it was found that slight changes in parameters and initial values may lead to obvious changes in the shape of attractors, so the change in the shape of attractors can be used to characterize the existence of small perturbations in the system. Then, the weak fault detection system based on the Chen oscillator was designed. According to the shape of the attractor of the Chen system, the shape of the system changes during the process of parameter change. A threshold selection method based on chaotic attractor shape step change was designed, which not only realizes automatic selection of the threshold, but also realizes the automatic recognition of the components to be tested in the input signal. Finally, the weak signal detection method based on the Chen system was applied to the early weak fault diagnosis of rolling bearings to judge the fault type. Simulation and experiments show that the method is effective and can accurately identify early weak faults of rolling bearings.