A test object can be selectively included in a test run based on predicting the behavior of the test object. In one embodiment, the present invention includes predicting how likely the test object is to produce a failure in a test run and deciding whether to include the test object in the test run based on the predicted likelihood. This likelihood of producing a failure may be based on any number of circumstances. For example, these circumstances may include the history of prior failures and/or the length of time since the test object was last included in a test run.