This chapter deals with the measurement of Value-at-Risk parameter for a portfolio using historical returns. The main issue here is the estimation of suitable percentile of the underlying return distribution. If returns were normal variates, the task would have been very simple. But it is well documented in the literature that financial market returns seldom follow normal distribution. So, one has to identify suitable distribution, mostly other than normal, for the returns and find out the percentile of the identified distribution. The class of non-normal distribution, however, is extremely wide and heterogeneous, and one faces a decision-making problem of identifying the best distributional form from such a wide class of potential alternatives. In order to simplify the task of handling non-normality while estimating VaR, we adopt the transformation-based approach used in Samanta (2003). The performance of the transformation-based approach is compared with two widely used VaR models. Empirical results are quite encouraging and identify the transformation-based approach as a useful and sensible alternative.