Abstract: Wind noise is the most common kind of noise in the nature, which will seriously reduce the quality of microphone recording. And because of its non-stationary characteristics, wind noise can not be suppressed by the common noise reduction methods, such as spectral subtraction and so on. In this paper, we analyze the spectral coherence of speech and wind noise from the dual-microphone, and then we take the idea of Zelinski post-filter to design a wind noise reduction filter, utilizing the strong coherence of speech and the weak coherence of wind noise between the two microphones. Also in the filter design, the combined effect of wind noise and background noise in the free sound field and the diffuse sound field is taken into consideration. The filter can detect the wind noise using the coherence of the signal and precisely estimate the wind noise coherence function. Experimental results show that the wind noise reduction method based on coherence of dual-microphone system significantly improves the denoising performance by contrast with other traditional noise reduction methods, also it has the characteristics of small computation and high real-time performance, which can be widely used in kinds of sound pickup systems.