This study makes a contribution to the literature by applying the Markov-Switching Bayesian VAR models for the first time to investigate the nonlinear linkage between gold prices and stock market index. Analyses have been done in the period from 1986: 04 to 2013:11. The Bayesian approach to econometrics provides a general method for combining modeller's beliefs with the evidence contained in the data. In contrast to the classical approach to estimate a set of parameters, Bayesian statistic presupposes a set of prior probabilities about the underlying parameters to be estimated. We use gold prices (USD/oz.) and S&P 500 Stock Price Index as an endogenous, the crude oil prices (Brent-$/barrel) as an exogenous variable in the analysis. We investigate the number of regime by LR test and The Markov Chain Monte Carlo (MCMC) algorithm and Sims & Zha (1998) prior distribution are employed to estimate the models.