The classic problem with the idea of mental representations and mental images.
The problem is usually stated like this: If you have an image in your mind, then who is it that is seeing the image? An internal image requires a "little man" or "homunculus" to look at it. And then the image in the head of this homunculus requires another homunculus to see it. And so on, creating a problematic infinite regress.
This may not seem like a problem in our our current intellectual paradigm, but it was, for a long time, seen as close to a complete refutation of the idea of mental images. Today there is a solution, coming from more holistic and less reductionist viewpoints. It has to do with the way we break down the mental processes going on.
The representational theory of mind has a good solution: Whatever the entity or mental module is that sees images in the head, it is stupider than the whole (the mind) of which it is a part. ("Stupid" here has a very technical definition -- something is stupid to the degree to which it lacks complexity). If this "image seeing" module requires other modules inside of it, they will have to be stupider still. Eventually, the process bottoms out with a module that is stupid enough to be self-sufficient and rigid.
Implicit in this solution is the idea that a complex, intelligent entity can be built entirely out of simple, stupid entities. This idea appears again and again when looking at intelligence and complexity. It is the relationships and connections, and not the entities themselves, that matter.