This study presents a fast orthogonal search (FOS) method for modeling fMRI time series. Based on the system identification theory, an orthogonalization procedure to model the fMRI time series is described. FOS method does not require equally space data, and can resolve sinusoidal frequencies much more closely than Fourier transform method. After the time series are modeled by means of the FOS, F-test is employed to detect the activation regions. Eight volunteers' data were collected to validate the proposed method. The results demonstrate the feasibility of the proposed method.