In order to monitor and identify transformer winding looseness fault more effectively, a chaotic feature analysis method for transformer winding looseness fault is proposed. Firstly, according to the chaotic dynamic characteristics of vibration signals, mutual information method and G-P algorithm are used to determine the delay time and embedding dimension respectively to reconstruct the phase space of transformer vibration signals; Secondly, the chaotic characteristic of transformer vibration signal is proved by judging whether the maximum Lyapunov exponent is positive. On this basis, the influence of different degrees of winding looseness fault on the change of phase space trajectory is analyzed; Finally, correlation dimension, Kolmogorov entropy and maximum Lyapunov exponent are used as a set of chaotic features to quantify the chaotic characteristics of vibration signals before and after the happening of transformer winding looseness fault. The results show that the maximum Lyapunov exponents of transformer vibration signals are all greater than 0, which proves that they have chaotic characteristics, and the obtained chaotic characteristics can effectively reflect the looseness fault of transformer windings. The research results provide a theoretical basis for monitoring the loosing state of transformer windings.