This paper examines if incorporating changes in financial market risk perception improves the predictive power of an early-warning system for systemic banking crises. In explaining systemic banking crises, the existing literature identifies inflating stock and real estate bubbles, credit booms, and surges in net capital inflows as the common drivers. Employing panel logit models to predict the postwar systemic banking crises in advanced economies to occur within three–four years, the paper’s key finding is that, even after controlling for the effects of surges in asset and credit markets and net capital inflows that are above the long-run trends for an extended period, market participants’ increasing underestimation of downside risks is a significant predictor of these crises. Incorporating changes in risk perception improves the prediction accuracy of the model significantly. This finding is robust across alternative prediction horizons, systemic crisis definitions, and risk-perception measures. Consistent with the recent theoretical developments in the form of the diagnostic expectations hypothesis for financial markets, the interpretation is that recent recurring good news about financial markets and the broader economic trends for sufficiently long periods lead to growing neglect of tail risks and riskier financial transactions, raising systemic risk and the likelihood of a financial crisis. The finding suggests monitoring financial market risk perception, in addition to the conventional indicators, to predict and avert systemic banking crises.