In real markets, trades occur discretely in time. Hence, bid-ask spread estimators that rely on the assumption of continuous trading are generally biased, understating transaction costs for less liquid assets. Here we formally derive an efficient estimator of the bid-ask spread when trading is discrete. The estimator is asymptotically unbiased and exploits the full set of open, high, low, and close prices to minimize the estimation variance. In absence of quote data, it delivers the most accurate estimates of bid-ask spreads theoretically, numerically, and empirically. The estimator is easy to calculate and has a broad applicability in empirical finance.