The global financial crisis pointed at the significant weaknesses of the existing models for measuring risk in financial markets. In particular, there is a need for improvement and further development of the financial risk models in the capital markets in transition countries. Within this paper we deal with the possibility of predicting and behavior of various classes of VaR models focusing on the capital market in Serbia from 2005 to the end of 2015.We implemented various VaR models in assessing market risk and used a two-stage backtesting process for VaR models validation. The results indicate that VaR based on symmetric GARCH model with generalized error distribution of innovations behaves reasonably well in the post-of-sample testing period. Filtered historical simulation at 99% confidence level gives the best results. Standard VaR models mostly used by financial institutions underestimate the risk forecast in the Serbian capital market in all market circumstances. Thus, the authors suggest to the regulators and investors to introduce more robust and complex risk measures in the Serbian capital market.