Backtesting evaluates model performance by applying it to historical data and measuring prediction errors. For VaR models, backtesting counts exceptions (days when losses exceeded VaR predictions) and performs statistical tests to assess if exception rates match confidence levels. Consistent underestimation indicates model failure requiring recalibration. Regulators require regular backtesting for market risk models used in capital calculations. A bank's daily VaR model with 99% confidence should experience exceptions on approximately 2-3 days per year; when backtesting revealed 12 exceptions, the model was recalibrated with updated volatility assumptions.