The vibration signal of wind turbine is a typical kind of signal with nonstationary and nonlinear properties. The ability of traditional methods to process the kind of signal is limited. In order to solve the shortage of the traditional methods and improve the ability of fault diagnosis, this paper proposes a new method to detect the fault based on the multifractal and support vector machine (SVM). Firstly, the multifractal spectrum of the input signal is calculated through the definition of multifractal. And then the eight fractal characteristics of signal are extracted. Finally, taking the characteristics as the input vector of SVM, it achieves the classification and recognition of normal signal and fault signals of four kinds of sun gears under different rotation speed. The experimental result confirms that the proposed method can effectively extract the characteristics of the planetary gearbox signal, and can raise the fault recognition rate under the condition of different rotational speed. This work is supported by National Natural Science Foundation of China (No. 51507098 and No. 51507100), Shanghai Talent Development Fund (No. 201365), Science and Technology Commission of Shanghai (No. 15YF1404600 and No. 13DZ2251900 and No. 10DZ2273400) and Shuguang Program (No. 15SG50).