Three experiments tested the hypothesis self-serving biases are self-deceptive in nature. Consistent with this hypothesis, Experiment 1 revealed that people who rated a task at which they succeeded as more important than a task at which they failed also cheated on a series of math problems, but only when they could excuse their cheating as unintentional. Experiment 2 replicated this finding and demonstrated that a self-report measure of self-deception did not predict this self-deceptive cheating. Experiment 3 replicated Experiments 1 and 2 and ruled out several alternative explanations. These experiments suggest that self-serving biases have a self-deceptive component, and that individual differences in self-deception can be measured.