Banks are required to maintain an appropriate level of capital which must commensurate with the riskiness of their portfolio. Recently, the Reserve Bank of India (RBI) issued a circular on Prudential Guidelines on Capital Adequacy-Implementation of Internal Models Approach (IMA) for Market Risk to select a suitable method for the banks to determine the regulatory capital requirement under the market risk exposure. Banks which adopt this approach are required to quantify market risk through their own Value-at-Risk (VaR) model. Therefore, it is a challenging task for risk managers of the bank to select an appropriate risk model which reasonably covers the risk of the bank's portfolio. Use of wrongly calibrated risk models may lead to undercapitalised banking system. This article aims at exploring the suitable risk model for measuring foreign exchange risk in banks' portfolio. The objective of present study is to empirically test the appropriate VaR model for foreign exchange rate risk. Value-at-Risk has been estimated for foreign exchange rate risk by using parametric variance-covariance method and non-parametric historical simulation (HS) method. Under parametric method, VaR is estimated by assuming that returns follow normal and Student's t-distribution. Backtesting results for various VaR models have been done based on Kupiec's proportion of failures (KPOF) test and regulatory 'traffic light' test. This article concludes that when returns are non-normal, VaR model based on the assumption of normality significantly underestimates the risk. Our empirical results based on backtesting show that most accurate VaR estimates are obtained from Student's t VaR model.