This study investigates whether there exist common model features that yield consistently superior results under both value-at-risk (VaR) and expected shortfall (ES) risk metrics in the energy commodities markets. We analyze the performance of ten VaR and seven ES models on the daily spot prices of West Texas Intermediate, Brent, natural gas, heating oil, US low sulphur coal and uranium yellow cake. Our backtesting results show that advanced semiparametric VaR and ES models, which entail sophisticated modeling of conditional volatility and extreme tails, are required to capture the true level of risk in the energy markets. We find consistency in the performance of tested risk models under both VaR and ES measures. In our tested sample, the filtered historical simulation and mirrored historical simulation models rank among the best VaR and ES models. A common feature of both models is that they do not make a priori parametrical assumptions about the return distribution but use empirical historical returns.