The harmonic current and vibration of the converter transformer are effective information for detecting winding mechanical faults. This paper proposes an online vibration frequency response analysis (VFRA) method and constructs a novel vibration frequency response function (VFRF), where the VFRF is defined as the ratio of vibration acceleration to current squared. A harmonic loading system simulating the actual operating situation of the converter transformer is established, and it is verified by numerical analysis and experimental methods that online VFRA is a simpler and more convenient method for winding mechanical condition detection than offline VFRA. The effects of different factors such as current magnitude and harmonic content on the online VFRA are investigated, and the converter transformer winding mechanical condition is detected by combining the VFRF trace with two numerical indices of relative factor and improved expectation. Laboratory experiments demonstrate that the online VFRA has high stability and successfully identify winding looseness faults of different degrees. Compared with the traditional frequency response analysis (FRA) and short-circuit impedance (SCI) methods, the online VFRA has higher sensitivity and detection capability for winding mechanical faults, and is a better method for online monitoring of winding mechanical condition.