Some things ought to be: for some A, it ought to be that A. Some things are better than other things: for some A, B, A is better than B. How can we spell out what ought to be in terms of what is better, or worse? There is an approach to this question which I will refer to as the standard one. (It is standard in the sense of being akin to a simplified core of what a number of authors say, and in validating the theorems of what is sometimes called standard deontic logic;' it is not standard in the sense of being received wisdom-as far as I can see nothing closely connected with deontic logic is received wisdom.) A's being better than B does not entail that it ought to be that A. Cheating is better than killing, nevertheless it is false that cheating ought to be. What ought to be is not what is better, but what is best. The standard semantics seeks to capture this by interpreting 'It ought to be that A'-O(A)-as true if and only if the best possible worlds are all A-worlds (worlds at which A is true). Similarly, the standard semantics interprets 'It ought to be that A given B'-O(A/B)-as true if and only if the best B-worlds are all A-worlds. I will assume some reasonable spelling out of the betterness relation between possible worlds without going into details. There are many ways in which this simple semantics can be modified, elaborated, and built upon. For instance, the possibility that there are no best worlds, or no best B-worlds, but rather an infinite series of ever better ones, can be covered by counting O(A) as true iff some A-world is better than any A-world, and by counting O(A/B) as true iff some AB-world is better than any AB-world. The possibility that the ranking of worlds is not absolute but varies from the perspective of different worlds, persons, or times can be covered by talking, not of the best worlds but of the best worlds from this or that perspective. The idea that in certain contexts the way things are in some worlds can be ignored-perhaps they are too remote from the actual world, or outside any agent's control-can be covered by counting O(A) as true in such contexts iff the best worlds accessible in that context are all A-worlds, and similarly for O(A/B). But the points we will be making