There are, first of all, some propositions which assert "relations of ideas "; there are, secondly, propositions which assert "matters of fact" which we ourselves are actually observing, or have observed and now remember; and there are, thirdly, propositions which assert "matters of fact" which we have never actually observed, but for believing in which we have some foundation in our past observations. And it is, I think, certain that some propositions, which we know as certainly as we know anything, do belong to each of these three classes. I know, for instance, that twice two are four; I know by direct observation that I am now seeing these words, that I am writing, and by memory that this afternoon I saw St. Paul's; and I know also that Julius Caesar was murdered, and I have some foundation in experience for this belief, though I did not myself witness the murder. Do any of those propositions, which we know as certainly as we know these and their like, _not_ belong to either of these three classes? Must we add a fourth class consisting of propositions which resemble the two last, in respect of the fact that they do assert "matters of fact," but which differ from them, in that we know them neither by direct observation nor by memory, nor yet as a result of previous observations? There may, perhaps, be such a fourth class; but, if there is, it is, I think, eminently desirable that it should be pointed out exactly what propositions they are which we do know in this way; and this, so far as I know, has never yet been done, at all clearly, by any philosopher. THE STATUS OF SENSE-DATA The term "sense-data" is ambiguous; and therefore I think I had better begin by trying to explain what the class of entities is whose status I propose to discuss.