Abstract: As a driver's state varies from normal driving to fatigue driving, the slowly changing and the rapidly changing waves, which are in his/her EEG(electroencephalogram), increase and decrease gradually, respectively. So, it is possible to detect fatigue driving based on the energy ratio of wavelet packet sub-bands. The EEGs of some subjects were acquired and analyzed during their simulating driving. The energy ratio of β waves to slowly changing waves was computed from the decomposition coefficients of wavelet packets and it was defined as a fatigue index F. Experimental result shows that F value is dependent on subjects, and there are big difference among the subjects' F values. But for any subject, his/her F value decreases slowly with the increase of driving time and fatigue level. The attenuation level of F value compared with that of normal state can reflect the driver's fatigue level.