A probabilistic seismic demand analysis method based on multivariate correlated kernel density estimation is proposed without any assumption on the distribution of engineering demand parameters（EDPs）. Three different correlation coefficients are used to describe the correlation respectively，and the traditional kernel density estimation is extended to multivariate correlated kernel density estimation based on the bandwidth matrix and multivariate Gaussian kernel function. A reinforced concrete frame shear wall structure is established based on SAP2000. The maximum inter?story drift ration and peak floor acceleration are selected to measure the multidimensional performance limit states. The probabilistic seismic demand model based on multivariate correlated kernel density estimation is established，and the structural demand annual average exceeding probability is obtained through Monte Carlo（MC）simulation. Traditional method based on multidimensional lognormal distribution assumption and the uncorrelated kernel density estimation are adopted for comparison. The research shows that：comparing with the traditional lognormal distribution assumption，the structure demand annual average exceeding probability based on the multivariate correlated kernel density estimation is larger，and the result of uncorrelated multivariate density estimation is smaller. Different correlation coefficients will affect the annual average exceeding probability，among which Pearson correlation coefficient is the most influential，Spearman correlation coefficient is the second，and Kendall correlation coefficient is the least.