The most distinctive prediction of prospect theory is the fourfold pattern (FFP) of risk attitudes. People are said to be (1) risk-seeking over low-probability gains, (2) risk-averse over low-probability losses, (3) risk-averse over high-probability gains, and (4) risk-seeking over high-probability losses. Using simple gambles over real payoffs, we conduct a direct test of this FFP prediction. We find that when pricing gambles subjects’ risk attitudes are consistent with the FFP. However, when they choose between the gamble and its expected value, their decisions are not distinguishable from random choice and are often the exact opposite of the prediction. These results hold both between and within subjects, and are robust even when we allow the subjects to simultaneously review and change their price and choice decisions.