This paper studies on the effect of kurtosis on fatigue life. A typical vibration test machine uses a normal distribution signal. However, the vibration signal generated in the actual operating environment is a non-normal distribution signal with kurtosis. In this paper, a signal which has the same power spectral density, but has different kurtosis is generated using the Zero Memory Non-Linear(ZMNL) method. The power spectral densities of these two signal will coincide within ±3dB. Fatigue damage spectrum(FDS) computed with a normal distribution signal and FDS computed with a non-normal distribution signal generated using the ZMNL method are compared. FDS computed using the normal distribution signal of kurtosis 3 will show a big difference over the 100Hz with FDS computed using the signal of kurtosis 6. This paper shows through such FDS comparison that a difference in fatigue life prediction may result in if a vibration profile which does not consider kurtosis occurring in the operating environment of a product is used in a vibration test.