Accurate identification of ferroresonance faults is a prerequisite for starting a microcomputer detuning device, but the existing time-frequency analysis methods are prone to misjudgment and mutual aliasing of adjacent frequency modes when extracting the characteristics of ferroresonance overvoltage signals. To solve the above problems, based on the chaotic characteristics of ferroresonance overvoltage time series, this paper presents a method for fault identification of ferromagnetic resonance in non-effectively grounded neutral systems based on phase space reconstruction. First, the characteristic parameters (embedding dimension and delay time) of phase space reconstruction are calculated by C-C method. Then, the overvoltage time series is introduced into the high dimensional space by phase space reconstruction technology to analyze the attribute features (correlation dimension and phase point distribution coefficient) and geometric features (radius vector offset and center distance) of the reconstructed attractor phase plane trajectory. The characteristics of reconstructed phase plane trajectory of different ferroresonance types are revealed. Finally, the K value is introduced as the characteristic parameter of change law of four kinds of characteristic value, and the reference K value of different types of ferromagnetic resonance is given in combination with the characteristic change law, which is used as the resonance criterion to identify the resonance type. The feasibility of the proposed method is verified by simulation analysis and a real ferromagnetic resonance case in a 10 kV distribution system in Yunnan province.