The invention discloses a method for determining an FMRI dynamic brain function time window. The method mainly comprises the following steps of obtaining functional magnetic resonance imaging data byusing magnetic resonance imaging and preprocessing the data, obtaining a single tested function connection network according to the sizes of 50TR, 100TR, 150TR and 195TR sliding time windows in the experiment, processing the obtained windows according to the sparsity of 0.05-0.5 (the step length is 0.05), and finally acquiring ten sparse matrixes for each window, solving small-world parameters forthe sparse matrix obtained by each window, solving a mean value and a variance, calculating a total mean value and a mean value of the variance, and comparing the attribute strength and robustness ofthe small-world network with different sliding time window sizes, and according to the attributes of the small-world network under different sliding time windows, using a similar binary search methodfor narrowing the range, and searching for an optimal sliding window. According to the method, a relatively reliable sliding time window in dynamic brain analysis is obtained by taking small-world parameter strength and attribute stability as standards, and support is provided for related research on brain function connection networks.