The reason why this is a fallacy may be understood by asking how the homunculus 'sees' the internal movie. The obvious answer is that there is another homunculus inside the first homunculus's 'head' or 'brain' looking at this 'movie'. But how does this homunculus see the 'outside world'? In order to answer this, we are forced to posit another homunculus inside this other homunculus's head and so forth. In other words, we are in a situation of infinite regress: the argument has failed. The problem with the homunculus argument is that it tries to account for a phenomenon in terms of the very phenomenon that it is supposed to explain--that is, homuncular arguments are fallacious for the same reason that a recipe for cake that had, as one of its ingredients, 'cake' is a problem recipe.
Another example is with cognitivist theories that argue that the human brain uses 'rules' to carry out operations (these rules often conceptualised as being like the algorithms of a computer program). For example, in his work of the '50s, '60s and '70s Noam Chomsky argued that (in the words of one of his books) human beings use Rules and Representations (or to be more specific, rules acting on representations) in order to cognise (more recently Chomsky has abandoned this view: c.f. the Minimalist Program).
Now, in terms of (say) chess, the players are given 'rules' (i.e. the rules of chess) to follow. So: who uses these rules? The answer is self-evident: the players of the game (of chess) use the rules: it's not the case (obviously) that the rules themselves play chess. The rules themselves are merely inert marks on paper until a human being reads, understands and uses them. But what about the 'rules' that are, allegedly, inside our head (brain)? Who reads, understands and uses them? Again, the implicit answer is (and, some would argue, must be) a 'homunculus': a little man who reads the rules and then gives orders to the body to act on them. But again we are in a situation of infinite regress, because this implies that the homunculus has cognitive process that are also rule bound, which presupposes another homunculus inside its head, and so on and so forth. Therefore, so the argument goes, theories of mind that imply or state explicitly that cognition is rule bound cannot be correct unless some way is found to 'ground' the regress.
This is important because it is often assumed in cognitive science that rules and algorithms are essentially the same: in other words, the theory that cognition is rule bound is often believed to imply that thought (cognition) is essentially, the manipulation of algorithms, and this is one of the key assumptions of some varieties of artificial intelligence.
Homunculus arguments are always fallacious unless some way can be found to 'ground' the regress (For example: a possible counter to this is that the brain as a whole is the homunculus, rather than thinking a specific part must be watching the movie). In psychology and philosophy of mind, 'homunculus arguments' (or the 'homunculus fallacies') are extremely useful for detecting where theories of mind fail or are incomplete.
The Homunculus fallacy is closely related to Ryle's Regress.