To what extent is the response to a consciously perceived word influenced by a prior word that is unconsciously perceived? Over the past 15 years or so, there has been a renewed interest in this question. Earlier research on this topic, dating back to the subliminal perception experiments popular in the 1960s, generated considerable controversy, and interest subsequently waned. Partly this was owing to fact that investigators relied on a form of priming that was not particularly robust (i.e., semantic/associative priming), and partly it resulted from profound skepticism about whether subjects were really unaware of the priming stimulus. Today the situation has changed completely: The phenomenon of masked priming has become routine, and masked priming techniques are applied to a wide variety of problems in visual word recognition, as the contents of this volume indicate. The reason for this was the discovery that repetition priming (as opposed to semantic or associative priming) was extremely robust under masking conditions (Evett & Humphreys, 1981; Forster & Davis, 1984), which meant that it survived under conditions where only the most narrow-minded skeptics would be tempted to argue that the subjects were really aware of the prime. So the question of whether unconscious perception was possible appears to have been dropped, and instead attention has been focused on the conditions under which masked priming effects could be observed. The aim of this chapter is first of all to survey some of the methodological issues involved in masked priming, and then to present new evidence about the mechanisms responsible for masked priming, mechanisms that operate at the earliest stages of visual word recognition, while processing is still controlled by orthographic form. The first experiment deals with the question of the site of priming. What is it that is primed? Does priming occur at the word level, or does it occur at a lower level? The second experiment examines the effect of the stimulus-onset asynchrony (SOA) between the prime and the target, and shows how this effect is relevant to models of priming. The last experiment considers whether it is possible to explain masked priming in terms of persisting actiuation, as one is forced to do in models such as the interactive activation model (McClelland & Rumelhart, 1981). Finally, we will take up a number of unresolved issues and suggest possible solutions.