In mechanical systems, their output performances depend on the statistical models of input variables. For correction estimation of output performance o mechanical systems, correct estimation of input models is also required. Especially, when input variables are correlated with each other, they need to be correctly modeled by multivariate density functions. However, the number of correlated data is very limited in real applications, and multivariate modeling of correlated variables is difficult using limited data. In this study, Bayesian method and multivariate kernel density estimation (MKDE) were used to estimate the multivariate density function, and sequential statistical modeling method (SSM) and Kernel density estimation with estimated bounded data (KDE-ebd) were used to estimate the marginal probability density function. Through the simulation tests, the multivariate statistical modeling methods were compared according to the various types of assumed true models. Engineering examples were used to evaluate the accuracy of various statistical modeling methods.