Well, in my limited experience, the best way to do testing is to start with tests first (TDD/BDD) - and the second best, when you've started, and have no tests - is to start from a bug: someone reports a real bug, in your real system - and you try and write a test that's failing in such a way, that when you alter the system to fix the bug, the test will pass.

Typically the first test will be quite painful - you might have to install a testing dependency, possibly re-work some code to allow for testing etc. But then, the next test should be a little easier. And after a few bugs and tests, you might find that for your next feature, you can somewhat easily start with a failing test.

/unsolicted_advice