In this paper we apply several non-parametric, parametric and semi-parametric methods in order to determine the market risk of Greek mutual funds, according to the Value at Risk measure. More specifically, the methods used are Historical Simulation for two sample sizes, Simple Moving Average and Exponentially Weighted Moving Average (RiskMetrics model), GARCH (1,1) and EGARCH (1,1) with normally and Student’s t distributed innovations, unconditional and GARCH (1,1) filtered Extreme Value Theory for different thresholds and Filtered Historical Simulation for two sample sizes. All calculations were made in a rolling basis and our sample consists of the logarithmic returns of nine mutual funds over the period 22/3/1993 to 21/11/2008 (3954 observations). The evaluation framework focuses on three tests proposed by Christoffersen (2003), namely unconditional coverage, independence and conditional coverage tests. Our results suggest that for the 95% VaR forecasts almost all methods do not perform well, while this is not the case for 99% confidence level. For the latter, some methods exhibit remarkably good performance, depending on the type of fund. Finally, we compute simple HS based Expected Shortfall just to give the reader an intuition of this alternative to VaR measure.