The cloud image method is widely used in the result vulnerability analysis because of its convenient calculation. However, there are some problems in the cloud image method, such as the engineering demand parameter (EDP) does not meet the lognormal distribution, the seismic intensity index ln(IM) does not meet the linear relationship with the engineering demand parameter (ln(EDP)), and the residual does not meet the normal distribution. Therefore, this paper introduces two methods of Box-Cox transform and 3-sigma criterion, and combines Latin hypercube sampling to propose a vulnerability analysis method that not only maintains the advantages of less analysis times of the cloud image method but also improves the normality of ln(EDP), the correlation between independent variables and dependent variables and the normality of residuals of cloud image method. Finally, taking a continuous beam bridge as an example, the time history analysis of the whole bridge model is carried out to verify whether the variables before and after the correction meet the normal distribution, and the effectiveness of the improved method is evaluated; By establishing the vulnerability curve of the cloud image method and the improved method, the cloud image method and the improved method are evaluated. The results show that the introduction of Box-Cox transformation and 3-sigma criterion can improve the linearity and normality of the probabilistic seismic demand model, and ensure the accuracy of the vulnerability calculation results. Using the Latin hypercube sampling method to consider the uncertainty of the structure can make the vulnerability results more realistic.