Empirical studies show that asset returns are non-normal and also time varying. These findings explain why the normality VaR and unconditional VaR perform poorly in market risk measurement and management. In the literature, the bivariate conditional copula function can be used to represent a joint distribution function of correlated asset returns whose marginal distributions can be of any form, and to accommodate time varying behavior of asset returns. This study extends the technique further to be able to measure the risk of a portfolio of any number of assets. The extension is successful because the study can endorse a positive definite correlation matrix of the conditional copula function by applying Semidefinite Programming (SDP). The study applies the method in risk measurement of Thai Government bond portfolios. In principle, the method should show improved performance over previous empirical studies, but the results from the Kupiec test show that the conditional copula method fails to estimate the VaR of these portfolios. This failure may have been caused by the misspecification of marginal distribution. [ABSTRACT FROM AUTHOR]