We will make our usual assumption: that TNT incorporates valid methods of reasoning, and therefore that TNT never has falsities for theorems. In other words, anything which is a theorem of TNT expresses a truth. So if G were a theorem, it would express a truth, namely: "G is not a theorem". The full force of its self-reference hits us. By being a theorem, G would have to be a falsity. Relying on our assumption that TNT never has falsities for theorems, we'd be forced to conclude that G is not a theorem. This is all right; it leaves us, however, with a lesser problem. Knowing that G is not a theorem, we'd have to concede that G expresses a truth. Here is a situation in which TNT doesn't live up to our expectations—we have found a string which expresses a true statement yet the string is not a theorem. And in our amazement, we shouldn't lose track of the fact that G has an arithmetical interpretation, too—which allows us to summarize our findings this way: A string of TNT has been found; it expresses, unambiguously, a statement about certain arithmetical properties of natural num bers; moreover, by reasoning outside the system we can determine not only that the statement is a true one, but also that the string fails to be a theorem of TNT. And thus, if we ask TNT whether the statement is true, TNT says neither yes nor no.