Basically, its a mathematical function where if you start at 0,0, you might falsely believe you are at the (or a) maximum or minimum of the function, as the slope at 0,0 is 0.
But, if you go any direction in the x axis, your function value rises, any direction in the y axis, your function value falls.
Thus a saddle point is an illusory, false impression of being at the extreme extent of a function, when in fact you are not.
The idea is that there is more to determining if you're truly at a global max or min of a function than only finding a single point where the slope is 0.
Edit: This comment if mine has been sitting at 69 upvotes for a few days now, so uh... obligatory joke about taste testing each others saddle points or something, I don't know.