To simplify the process of selecting ground motion intensity measures and further reduce the variability of probability seismic demand models without significantly increasing computational cost, a vector-valued intensity measure selection model for specific site conditions and structure types was established. First, based on 20 scalar-valued intensity measures (IMs) candidates, the variance inflation factor method was used to eliminate the unimportant IM which can be linearly represented by other IMs, thereby eliminating the collinearity among IMs. Then, non-collinearity IMs which have significant correlation with engineering demand parameters were picked out as components of vector-valued intensity measure based on partial determination coefficient. Meanwhile, a three-span prestressed concrete continuous girder bridge, located on the east Nanhuan Road of Handan city, was taken as an example to verify the validity of the proposed method. Results of 6-fold cross-validation show that the variability of the regression model established by vector-valued IM [S_a(0.2),S_a(1.0),T_m] on training dataset and test dataset was reduced by 13.67% and 17.07% respectively, compared with the regression model established by scalar-valued IM[S_a(1.0)], and no overfitting phenomenon was observed. The research results can be utilized to select appropriate vector-valued intensity measures for continuous girder bridges under specific site conditions.