Major empirical regularities in the distribution of time series of financial asset return data are their skewness and excess kurtosis. To capture the non-normality in estimating financial time series using GARCH models, many researchers have introduced more flexible parametric distributions than the normal distribution such as Student-t and GED. While considerable attempts have been made in modeling conditional distribution in univariate GARCH models, there has been little work on the multivariate conditional distribution function in multivariate GARCH models. Most academic studies in multivariate GARCH models have relied on multivariate normal distribution. The purpose of this paper is to introduce into the multivariate GARCH models a flexible multivariate parametric distribution in order to capture the non-normality of financial time series in a multivariate dimension. For an illustrative purpose, we apply the new bivariate distribution function to modeling time varying dependence of returns on KOSPI and won/dollar exchange rate, and compare the results with those of normal and Student-t models.