A method of statistical modeling of spatial structures based complex textural image pattern recognition is presented. The Weibull distribution process of the spatial structures in the complex textural images is analyzed theoretically in advance. Subsequently, by filtering the images with a proposed filter bank of Gaussian derivative filters of multi-scale and omnidirectional, the multi-scale omnidirectional spatial structure features of the complex textural images are obtained by statistical modeling of the image’s spatial structures. Based on the principle of partial least squares discriminant analysis, the image texture classification recognition model is established, which can be used to identify the pattern of the complex texture images effectively. Experimental results show that, the proposed image statistical modeling method achieves distinctive visual perception characteristics of the complex textural images, which has strong classification ability and stable classification performance.