We propose an empirically relevant measure of voting power that uses the information about real or assumed voting patterns conveyed by a joint probability distribution on the set of voting outcomes, and apply it to the voting data of the Supreme Court of the United States.