Suppose the astronomers discover that the stars do not exactly obey Newton's law. They will have the choice between two attitudes; they may say that gravitation does not vary exactly as the inverse of the square of the distance, or else they may say that gravitation is not the only force which acts on the stars and that there is in addition a different sort of force. In the second case, Newton's law will be considered as the definition of gravitation. This will be the nominalist attitude. The choice between the two attitudes is free, and is made from considerations of convenience, though these considerations are most often so strong that there remains practically little of this freedom. We can break up this proposition: (1) The stars obey Newton's law, into two others; (2) gravitation obeys Newton's law; (3) gravitation is the only force acting on the stars. In this case proposition (2) is no longer anything but a definition and is beyond the test of experiment; but then it will be on proposition (3) that this check can be exercised. This is indeed necessary, since the resulting proposition (1) predicts verifiable facts in the rough. It is thanks to these artifices that by an unconscious nominalism the scientists have elevated above the laws what they call principles. When a law has received a sufficient confirmation from experiment, we may adopt two attitudes: either we may leave this law in the fray; it will then remain subjected to an incessant revision, which without any doubt will end by demonstrating that it is only approximative. Or else we may elevate it into a _principle_ by adopting conventions such that the proposition may be certainly true. For that the procedure is always the same. The primitive law enunciated a relation between two facts in the rough, _A_ and _B_; between these two crude facts is introduced an abstract intermediary _C_, more or less fictitious (such was in the preceding example the impalpable entity, gravitation). And then we have a relation between _A_ and _C_ that we may suppose rigorous and which is the _principle_; and another between _C_ and _B_ which remains a _law_ subject to revision. The principle, henceforth crystallized, so to speak, is no longer subject to the test of experiment. It is not true or false, it is convenient. Great advantages have often been found in proceeding in that way, but it is clear that if _all_ the laws had been transformed into principles _nothing_ would be left of science. Every law may be broken up into a principle and a law, but thereby it is very clear that, however far this partition be pushed, there will always remain laws. Nominalism has therefore limits, and this is what one might fail to recognize if one took to the very letter M. LeRoy's assertions. A rapid review of the sciences will make us comprehend better what are these limits. The nominalist attitude is justified only when it is convenient; when is it so?