In this paper, wavelet transform and clustering method are combined in order to detect the task stimulation-caused activation areas of a human brain in fMRI. According to the idea of multi-step decision, firstly, we perform fuzzy-c-means clustering in the fMRI image to solve the ill-balanced problem of the data. Secondly, we use the stationary wavelet transform to decompose the data, and extract those data of interested frequency, and then the improved k-means clustering algorithm is proposed to analyze these data in the wavelet domain. The experimental results with several subjects show that the visual stimulation-caused activation areas of human brain can be detected. Compared with the popular SPM method, the proposed method in this paper is more reasonable, and has directive significance and practical value on the functional connectivity detection of human brains.