In the contemporary philosophical discussion of self-deception, the process of defining the concept is often conflated with explaining it. This results in much contention and confusion. To avoid both, I keep the two projects separate. In particular, I focus on the methods used to define self-deception. I base much of my work on the only extended discussion of the methods used to define self-deception given by Alfred Mele. It turns out, though, that all of Mele's methods have the same problem; the resulting notion of self-deception balloons out to encompass different mental phenomena that are too diverse. In order to narrow the definition of self-deception, I modify one of Mele's methods. Self-deception should be defined, I argue, by joining the concepts of self and deception. The resulting picture of self-deception is that it is an intentional deceptive act that a self attempts on itself. However, many hold that those who believe that self-deception is an intentional act of this sort must resolve two difficult puzzles. This is accepted both by those who support and by those who oppose the intentionalist picture of self-deception. I break this trend and accept the result of one of the puzzles: the process of intentional self-deception must undermine itself. Interestingly, taking on this position does not make self-deception paradoxical; a self-deceiver is free to continue to try to trick him- or herself even though it is impossible to succeed at this. I argue that this view respects the notion that a self-deceiver is literally attempting to trick him- or herself while avoiding the common worries that plague other intentionalist views. The inspiration for this view comes from Jean-Paul Sartre's discussion of bad faith. By first analyzing the process of defining self-deception and then using that process, I hope to have avoided the tendency of philosophers working on the topic to talk past each other. The considerations I have used to define self-deception are explicitly laid out and so too are the reasons why my definition should be preferred.