What conditions, then, must observations satisfy, if they are to justify a generalisation? Let us suppose that the generalisation to be justified is one which asserts that the existence of a kind of object, which we will call A, is generally preceded, accompanied, or followed by the existence of a kind of object, which we call B. A, for instance, might be the hearing of a certain word by one person, and B the thought of that which we call the meaning of the word, in another person; and the generalisation to be justified might be that when one person hears a word, not spoken by himself, someone else has generally thought of the meaning of that word. What must I have observed, if the generalisation that the existence of A is generally preceded by the existence of B, is to be justified by my observations? One first point, I think, is plain. I must have observed both some object, which is in some respects like A, and which I will call _α_, and also some object in some respects like B which I will call _β_: I must have observed both _α_ and _β_, and also I must have observed _β_ preceding _α._ This, at least, I must have observed. But I do not pretend to say _how_ like _α_ and _β_ must be to A and B; nor do I pretend to say how often I must have observed _β_ preceding _α_, although it is generally held that I must have observed this more than once. These are questions, which would have to be discussed if we were trying to discover what observations were _sufficient_ to justify the generalisation that the existence of A is generally preceded by that of B. But I am only trying to lay down the minimum which is _necessary_ to justify this generalisation; and therefore I am content to say that we must have observed something more or less like B preceding something more or less like A, at least once.