A classic result by Merton (1973) is that, except just before expiration or dividend payments, one should never exercise a call option and never convert a convertible bond. We show theoretically that this result is overturned when investors face frictions. Early option exercise can be optimal when it reduces short-sale costs, transaction costs, or funding costs. We provide consistent empirical evidence, documenting billions of dollars of early exercise for options and convertible bonds using unique data on actual exercise decisions and frictions. Our model can explain as much as 98% of early exercises by market makers and 67% by customers.