No, neither should UPDATE ordering. And (to your previous question): No, calling it a bug is not a smart conclusion - hope that wasn't you or a loved one! :) ORDER BY is an output modifier not a relational method. Once you dictate in which order the engine should update or delete entries, you remove from it any optimization opportunity, or possibility thereof. You attempt to interfere with the engine process and no longer simply the veracity of the data or statements. You neuter the QP - undermining the very utility of an RDBMS engine. Further, I would always want the feedback to be in order of actual deletes, but have space for people needing a different order, however that can easily be done. What about stipulating the order of UPDATEs or DELETEs? Well, IF that is really really needed, you can easily achieve that by simply stating exactly which item to UPDATE or DELETE, and then the next, rinse, repeat, in any order you want it to happen, whether writing an SQL script or doing it in your own code. I can't see a situation in which you would want to UPDATE or DELETE whole blobs of records, but then also needing to fine-grain control exactly which of them go first, I mean, not unless you are using a severely broken relational DB schema. As to the other reply specifying it MUST be so because of Postgres - Well, its outcome might agree with mine, but it's an argument from authority fallacy to start with, even though a really good authority in this case. I feel looking into WHY Postgres did it, and holding up that point would be a better motivated argument.