Among the legacies of Elizabeth Anscombe's 1957 monograph Intention are the introduction of the notion of ‘practical knowledge’ into contemporary philosophical discussion of action, and her claim, pursued throughout the book, that an agent's knowledge of what he is doing is characteristically not based on observation. Each idea by itself has its own obscurities, of course, but my focus here will be on the relation between the two ideas, how it is that the discussion of action may lead us to speak of non-observational knowledge at all, and how this notion can be part of the understanding of a kind of ordinary knowledge that we have reason to consider practical rather than speculative. Anscombe mentions several quite different things under the heading of ‘non-observational knowledge’, and she first introduces the notion of the nonobservational for purely dialectical purposes, associated with the task of setting out the field she wants to investigate, in a way that avoids begging the very questions she means to raise. She needs a way of distinguishing the class of movements to which a special sense of the question ‘Why?’ applies, but which doesn't itself employ the concepts of ‘being intentional’ or ‘acting for a reason’. Section 8 begins: “What is required is to describe this class without using any notions like ‘intended’ or ‘willed’ or ‘voluntary’ and ‘involuntary’. This can be done as follows: we first point out a particular class of things which are true of a man: namely the class of things which he knows without observation.” (p. 13) She first illustrates this by the example of knowledge of the position of one's limbs, the immediate way one can normally tell, e.g., whether one's knee is bent or not. But examples of this sort are in fact ill suited to shed light on the idea of ‘practical knowledge’, which is the true focus of the idea of the non-observational in the study of action. When we see this we will be better able to see why Anscombe is concerned with the non-observational in the first place, and how this concern is tied to other characteristic Anscombian theses, for instance that an action will be intentional under some descriptions but not others, and that practical knowledge is distinguished from speculative knowledge in being “the cause of what it understands” (p. 87). And we will be able to understand how it is that an agent can be said to know without observation that he is doing something like painting the wall yellow, when this knowledge so patently involves claims about what is happening in the world, matters which it seems could only be known observationally.