Abstract: Seismic hazard analysis and seismic fragility analysis are import parts of seismic risk analysis of the nuclear power plant, and the site-specific spectra generation and selection of ground motion records are the key connecting steps between seismic hazard analysis and seismic fragility analysis. The site-specific spectra based on scalar probabilistic seismic hazard analysis and seismic hazard disaggregation have been widely studied and used. In order to overcome some shortcomings of some site-specific spectra such as uniform hazard spectrum and conditional mean spectrum, the theory of generalized conditional mean spectrum with multiple conditional periods has been proposed. However, the generalized conditional spectra based on the characteristics of Chinese seismicity have not been studied as far as the author knows. Meanwhile, the safety factor method is a common method for seismic fragility analysis of the nuclear power plant. In recent years, the fragility analysis of the nuclear power plant based on analytical fragility data has been applied and studied. At present, the seismic fragility of the nuclear power plant has been usually analyzed based on one single ground motion intensity measure, which contains limited information, and the obtained fragility analysis results are uncertain. The research on multivariate fragility of the nuclear power plant based on analytical seismic fragility analysis methods such as incremental dynamic analysis is still lacking. In this paper, the multivariate seismic fragility analysis of a nuclear power plant containment in China was conducted via the generalized conditional spectrum for the nuclear power plant site in China. Firstly, based on the vector probabilistic seismic hazard analysis and disaggregation theory of the site in China, the generation method and steps of the generalized conditional spectra of the site in China were proposed. For the example site located in China, the generalized conditional spectra of the example site were generated. Based on the generated generalized conditional spectra of the example site, the ground motion records were selected via greedy optimization algorithm, and based on the multivariate seismic fragility analysis method, the multivariate seismic fragility surface of the containment structure of the example site were generated. The seismic fragility analysis results of the nuclear power plant containment are sensitive to multiple ground motion intensity measures. Based on analytical seismic fragility methods such as incremental dynamic analysis, more refined fragility analysis results can be obtained. Seismic fragility analysis results considering multiple ground motion intensity measures can provide research basis for more refined seismic risk of nuclear power plants.