Automaticity is an important phenomenon in everyday mental life. Many routine activities are performed quickly and effortlessly with little thought, in short, automatically. Over the last decade, psychologists have learned a lot about its empirical characteristics, but automaticity is poorly understood theoretically. This research addresses a new theory of automaticity, testing its assumptions and comparing its predictions against previous approaches. The theory assumes that automatic processing is based on direct-access, single-step memory retrieval, as when 4 pops into mind as the sum of 2+2 or the route home comes to mind as we enter our cars at the end of the day. Practice is important because it strengthens memory and makes it more reliable, so that performance on a well-practiced task can depend entirely on memory retrieval. The assumptions about strengthening memory allow the theory to predict the form of the learning curve and the reduction of verifiability with practice, which are basic effects that all theories must predict. The research examines the role of attention in encoding into memory and retrieval from memory. The theory assumes that whatever is attended will be encoded into memory, which implies certain context effects not predicted by other theories. Specifically, it implies that memory will contain a trace of the person's attentional history with a stimulus, associating the different things the person attended to in responding to the stimulus. That trace will be different depending on the attentional requirements of the task. Tasks in which attention can be focused directly on the stimulus will have weaker context effects than tasks in which the person must search for the stimulus, dividing attention among several possible stimuli. Previous approaches predict no effect of context regardless of the conditions of attention. The theory assumes that attention to a stimulus causes whatever was associated with it in the past to be retrieved. This is the basis of the speed and effortlessness of automatic processing. In addition, it implies that knowledge about the stimulus is available only after the person attends to it. Thus, directing attention toward a familiar stimulus should "turn on" automatic processing and directing attention away from it should "turn it off." This contrasts with previous approaches, which assume that automatic processing becomes available prior to attention, before attention is focused on the stimulus.