(3) The conception of an object or an action proves that it existed, but not that it was common; the object or action may have been unique, or restricted to a very small circle; poets and novelists are fond of taking their models from an exceptional world. (4) The facts yielded by this method are not localised in space or time; the author may have taken them from a time or country not his own. All these restrictions may be summarised as follows: before drawing any inference from a work of literature as to the state of the society in which the author lived, we should ask ourselves what would be the worth of a similar inference as to contemporary manners drawn from a modern novel. With the facts yielded by conceptions we may join those indifferent facts of an obvious and elementary character which the author has stated almost without thinking. Logically we have no right to call them certain, for we do sometimes meet with men who make mistakes about obvious and elementary facts, and others who lie even on indifferent matters. But such cases are so rare that there is not much danger in admitting as certain facts of this kind which are supported by a single document, and this is how we deal, in practice, with periods of which little is known. The institutions of the Gauls and Germans are described from the unique texts of Cæsar and Tacitus. Facts so easy to discover are forced upon the authors of descriptions much as realities are forced upon poets. II. On the other hand, a statement in a document as to an objective fact is never enough to establish that fact. The chances of falsehood or error are so many, the conditions which gave rise to the statement are so little known, that we cannot be sure that none of these chances has taken effect. The critical examination provides no definitive solution; it is indispensable if we are to avoid error, but it is insufficient to conduct us to truth. Criticism can _prove_ no fact; it only yields probabilities. Its end and result is to decompose documents into statements, each labelled with an estimate of its value--worthless statement, statement open to suspicion (strong or weak), statement probably (or, very probably) true, statement of unknown value.