An event-driven backtest is a form of backtesting in which a simulator feeds historical data to a strategy one event at a time, in strict time order. Each new price, order acknowledgment or fill is an event. The strategy reacts to it, may send simulated orders, and those orders are filled by a model of the market before the next event is processed. The design follows the event-driven architecture of real trading systems, and the same strategy code can often run unchanged in live trading.
The alternative is a vectorized research backtest, which computes signals and returns for the whole history at once. That method is fast but crude about execution. An event-driven test is slower to build and run, yet it can model order types, queue position, partial fills, slippage and delays, and it makes look-ahead bias harder to introduce, because the strategy never sees data beyond the current event.