I love the following definition of fun, as I find it matches perfectly my own game experiences. You can find the original reference here, with link to the MSc thesis of the person making this conjecture:
[…] enjoyment is based more on proximity to a performance “sweet spot” just past the win/loss threshold rather than on learning effects. Combining my results with Kevin Burns’ information-theoretic model for enjoyment , I conject that “fun” is based on goal-satisfaction under the conditions that the desired outcome is uncertain.
This is very important to know when designing a game. MUDs I use to play come to my mind: even though they were technically simple, they managed to hook generations of kids (and “less kids”) for years, hours over hours every day, just because they wanted to reach a final goal (being the most powerful, usually, but also social power in certain cases, or ascending to “Immortal-level”, ie becoming part of the MUD staff).
That was only one side of it of course: whenever the game administators decided to make certain goals more accessible (for example by probabilistically uniforming all combat styles, such that whatever style you choose would bring you, eventually, to perfect parity with any other opponent), usually many players stopped having fun and left. Same goes for promoting people to staff level too freely: they abandoned the game quickly after.
It’s a good lesson, one many game administrators know by heart maybe, but repetita iuvant doesn’t it?