Abstract: In order to minimize the structural response variation caused by epistemic uncertainty within the ground motion input, it is required that the selected ground motion records should reflect the seismic hazard level of the engineering site. The traditional ground motion selection method based on conditional mean spectrum theory is greatly influenced by the choice of conditional ground motion intensity measures. In this study, multivariate hazard consistent target spectrum is constructed based on the concept of multivariate return period, which is applicable to China considering the characteristics of the three-level seismicity source and the long and short-axis ground motion prediction equation in probabilistic seismic hazard analysis. Taking a site in Tianjin as an example case, the ground motion selection results under the definition of multivariate return period are compared, including joint return period, co-occurrence return period and Kendall return period. Taking the multi-degree-of-freedom shear model and the high-rise building significantly affected by multi-model effect as examples, the differences in the structural seismic response under the selected ground motion records following the multivariate hazard-consistent target spectrum and the traditional conditional target spectrum were compared. The results show that the hazard-consistent target mean spectrum defined by the Kendall return period lies between the conditional target spectra of different conditional periods, while the target spectrum defined by the co-occurrence return period and the joint return period are overly conservative or underestimated. When the return period is long and the ground motion intensity is high, the structural response under the traditional conditional spectrum is greatly influenced by the choice of conditional period. While the Kendall hazard-consistent target spectrum is not sensitive to the choice of conditional period, and the overall spectrum shows good stability. Based on the ground motion selection results, the structural response is not significantly affected by the choice of conditional period, and a stable median value estimate of the response can be obtained. However, the multivariate hazard-consistent target spectra proposed in this study use multivariate return periods as the target to control the spectrum shape, which can well solve the shortcomings that the traditional conditional spectrum requires manual determination of the anchoring conditional ground motion intensity measures, and can be applied to performance-based seismic risk analysis of structures.