A modified method is proposed for the probabilistic seismic hazard analysis considering nonlinear site effect. A case study is presented to show the hazard curves that consider the nonlinearity of the soil layers at Gilroy #2. Source parameters and an earthquake occurrence model are assigned to the Caleveras fault, which is assumed the dominant seismic source to the site. A relationship between ground surface and bedrock accelerations at the site is established by recordings from three earthquake events using the program NONLI3, which calculates the nonlinear responses of the soil layers to strong motions. Various values of standard errors of the estimated surface PGA, given the bedrock motions, are assumed to perform a sensitivity analysis. An alternative nonlinear relation is included in the case study. In addition to the nonlinear models, a linear-response model is also performed using the same soil profile at Gilroy #2. The results demonstrate that the hazard curves are sensitive to the standard error of the estimated surface PGA values at the site. The hazard curve may be severely distorted if the soil condition of a nonlinear site is treated inadequately. The alternative nonlinear model obtained empirically from regression analysis and the linear response model are both unsuitable for the hazard assessment considering nonlinear site effect.