If you want activated regions only, then you should apply a t-test (F tests find differences in 'any' direction while t-tests distinguish between directions).

There are some predefined contrasts in SPM when you estimate a model (e.g. the "effects of interest" f-contrast). If you hadn't defined your control (or baseline) task then it would find areas activated for stimulus A and also areas activated for stimulus B.