Abstract: Recently, a growing number of studies have shown that machine learning technologies can be used to decode mental state from functional magnetic resonance imaging (fMRI) data. It has achieved the functional positioning of the brain activity using fMRI technology to interpret the thinking data.But how to run the specific operation of the brains thinking process is still unknown. The analytical methods to better reveal the thinking process need to be further studied. Through adopting two stimuli tasks of solving the 4×4 Sudoku problems and image emotional reaction, the thinking process data which interpret the different state of mind is gotten, and different ways of thinking data analysis methods are explored. The experiments proved that the feature selection methods of t-test, the feature extraction methods of the peak time and the classification algorithm of SVM are more suitable for analyzing the fMRI data, especially to two different states of mind data above, which can reveal the correct state of mind. This essay should provide the theory basis and data for promoting the fMRI technology to interpret the thinking data.