This thesis attempts to outline a solution to the general problem of how organisms control their behaviour relative to the environment. The principal concern is with how such guidance is achieved on the basis of visual information. The main argument is that behaviour is controlled on the basis not of current information but of previously acquired information. Experiments are reported which show that even complex locomotor acts can be accurately executed when all visual information is excluded during the act. The research suggests two ways by which motor activity is controlled. Firstly, visual information can be used directly to formulate programs for action. These programs contain precise prescriptions of the motor actions necessary for controlling the behaviour, and are executed with minimal reliance on information available during the act. It seems that programs can be formulated for distances up to about five metres. Secondly, a map or internal representation of the external environment can be formed which can be used in place of direct vision to formulate programs and control behaviour. Such maps allow behaviour to be controlled over distances up to 21 metres at least, though they are an effective substitute for vision for not more than about eight seconds. Programs seem to have a longer duration. The findings are taken to imply that vision is used to a far lesser degree in on-going control than is commonly supposed. Two basic reasons are proposed for this. Firstly, it is argued that only a map/program strategy of this kind can allow the complexity typically seen in visually-guided behaviour. Secondly, vision must frequently be free to perform its many other functions. Only by devolving its work-load can it achieve this necessary freedom.