In order to improve the poor prediction precision of fatigue life under a wide-band random load, a Weibull model with varied parameters is established to describe the peak value probability density without correcting the probability density error function. The fatigue life is predicted. The theoretical probability density of the peak value is first given and converted from Rayleigh distribution to Gauss distribution using the Weibull model with varied parameters. The range of the model parameters is also discussed. The theoretical expression of the prediction model of fatigue life is obtained according to the Weibull model with varied parameters. The fatigue lives under a wide-band random load with double peaks and under a continuous wide-band random load are calculated, and the accuracy of various models is compared. The results show that the theoretical error of the proposed modelis 9.68% and the experimental error is 12.5%, which satisfies engineering needs. In addition, the proposed method has a more convenient application compared withthe Dirlik method.