We develop a transactions-based classification model to identify retail investors’ transactions by using a definitive identification of investor types as issued by the South Korean exchange. Logistic regressions and Random Forest models are trained with known investor types and classify daily retail trades on anonymized out-of-sample transactions. Both models deliver about twice higher classification accuracy than existing methods, particularly for small-cap stocks. Classification accuracy declines systematically with firm size. Because retail investors prefer small stocks, and illiquid stocks in all size quintiles, trade-based identification of retail trading activity is most reliable in small-cap and illiquid stocks.