I thought this was a no brainer until I saw that 37% got it wrong. It always use to bug me when one of my previous team leaders would talk about "compiling a stored procedure" when he would issue a create/alter. Performance tuning is a related, but descrete discipline unto itself.

Anyway, the only thing I would add to the explanation on your post is that the optimizer may decide to recompile the plan on subsequent runs if the statistics on the underlying tables change significantly. Other possibilites that could cause the plan to be recompiled upon execution are temp tables and the with recompile option.

Regards,

Toby