Capacitive voltage transformers (CVTs) are important voltage state monitoring equipment in substations, and their safety and stability are related to the safe operation of the power grid. Disintegrating CVTs for fault diagnosis can cause damage to equipment. When CVTs are not disintegrated, their latent failure is difficult to evaluate accurately, but simply collecting frequency electrical quantities is insufficient when diagnosing typical internal faults. In this study, a fault diagnosis method based on the broadband transfer characteristics of CVTs is proposed for the three most common faults: short-circuited high-voltage capacitive elements in CVTs, short-circuited medium-voltage capacitive elements, and short-circuited primary winding of intermediate transformers. The network analyzer was used to test the port scattering parameter matrix of the CVT under normal and fault conditions and convert it into a voltage transfer function. Then, the resonance point corresponding to the first peak on the broadband voltage transfer function curve was selected as the characteristic point, and the different types of faults and different degrees of faults were analyzed. Finally, the magnitude gain offset, offset intensity, and phase offset of the normalized voltage transfer function at the resonant point were introduced, and the influence of CVT voltage transfer characteristics was used to quantify the fault state of the CVT. On this basis, fault diagnosis criteria were utilized to realize the nondisintegration diagnosis of the typical internal faults of the CVT.