Systematic overconfidence by individuals regarding their abilities and prospects could have important economic consequences. But overconfidence has received little direct testing within economics. We use experiments to test for overconfidence in people's forecasts of their absolute or relative performance in two unfamiliar tasks. Given their chosen effort, participants have incentives to forecast accurately, with opportunities for feedback, learning and revision. Forecasts are evaluated at aggregate and individual levels. We find zero mean error or underconfidence far more prevalent than overconfidence. Underconfidence is greatest in forecasts of absolute rather than relative performance and among those using greater effort quantity or quality.