A broad discussion on the representational problems of cognitive science is presented. Two common classes of representations, symbol based propositional representations and input-output mappings referred to as weak representations, are described and their relation to the physical symbol system (PSS) and connectionist approaches (CA) is discussed. Four problems often used as critique of PSS, the ontological assumption, the frame problem, the symbol grounding problem and the frame of reference, is shortly presented and it is concluded that these problems to some extent also apply to many connectionist models. Embodied cognitive science is presented and it is argued that weak representations used in embodied agents avoid the four fundamental problems, but are instead subject to the problem of adaptation. It appears that, in order to solve the problem of adaptation, some bias for generalization has to be introduced, which also would reintroduce the problems with propositional representations. The conclusion is that the fruitful solution to the fundamental representational problems of cognitive science is to focus on these problems with a joint view of propositional and weak representations, rather than trying to rule out one in favor for the other.