One of the most widely used methods to quantify risk is ‘Value at Risk’. VaR models are useful only if they predict future risks accurately. This paper focuses on a comparative evaluation of three broad approaches to calculate VaR for nine commodities traded on Multi Commodity Exchange of India. The primary objective of the study is to identify the most accurate VaR model for each commodity in particular and commodity asset class in general. VaR is calculated using five different methods (two methods each of parametric & non-parametric approaches and one method of semi-parametric approach) for all nine commodities for a period of nine years starting October 2006 till October 2015. To identify the better performing VaR methods accurately, the analysis is performed in two phases, Pre-Crisis (October 2006 to December 2009) and Post Crisis (January 2010 to October 2015). Results suggest Volatility Weighted Historical Simulation (VWHS) VaR method has outperformed other methods in both parts of the analysis exhibiting a success ratio of 100% each time. We also conclude that the selection of similar or contrasting data periods in terms of market conditions for VaR calculation and VaR backtesting affects the performance of VaR methods in general. These findings are relevant for retail and institutional investors who hold commodities in their portfolios and traders who need to calculate VaR for their commodity portfolios.