How do you choose the right cognitive task for a study?

I keep running into the problem that several tasks seem to tap the same process, but each has different confounds and norms.

I usually end up picking the one with the best validation data, but I'm never fully confident.

How do you decide which task to use when you're designing a new study?