Ideally I want to do this:
UPDATE TOP (10) messages SET status=10 WHERE status=0 ORDER BY priority DESC;
In English: I want to get the top
I have to offer this as a better approach - you don't always have the luxury of an identity field:
UPDATE m
SET [status]=10
FROM (
Select TOP (10) *
FROM messages
WHERE [status]=0
ORDER BY [priority] DESC
) m
You can also make the sub-query as complicated as you want - joining multiple tables, etc...
Why is this better? It does not rely on the presence of an identity field (or any other unique column) in the messages table. It can be used to update the top N rows from any table, even if that table has no unique key at all.