Multivariate stochastic volatility (MSV) models have been intensively studied in the past several years. For the general case of the MSV models, there are not many methods having been proposed. In this talk, correlations are permitted between the innovations of the asset returns and those of the volatility dynamics. We look at the MSV model in a Bayesian framework by applying an inverse Wishart distribution. The multistage slice sampler within the Gibbs algorithm is proposed to sample the persistent parameters and latent variables. Since the Metropolis-Hastings (M-H) method is avoided, our algorithm is more efficient and easier to operate.