A new theory of visuo-spatial mental imagery and a computational model of the theory are presented. The theory assumes (visual) perception to be an active and guided process comprising of several low-level perceptual actions. Mental concepts are grounded in sets of such actions. Imagery comprises the "offline" employment of these actions providing concrete instances of the mental concepts through bodily feedback such as proprioception. The theory is compared to the contemporary theories and evaluated against a set of well established phenomena, i.e., mental scanning, mental reinterpretation, eye movements, unilateral neglect. It is argued that the theory provides explanations that go beyond those offered by the contemporary theories. The results provide support for the explanatory power of an embodied approach to cognition in which sensorimotor interaction constitutes conceptual knowledge in that it provides grounding and concrete manifestation of the semantics of mental concepts in the domain of visuo-spatial mental imagery.