Numerous investigators have reported difficulty obtaining reliable learned helplessness. Various laboratories have used differing test environments and criteria, making comparisons among experiments difficult. Some use an escape deficit criterion, in which escape is slowed down in a shuttle box, while others have used an escape failure criterion, in which rats do not escape at all on most test trials. Little work has been done to test the validity of LH, i.e., the prediction of persistence of escape failure after exposure to uncontrollable shock. The present studies demonstrate that the reliability and validity of learned helplessness can be improved by 1) modifying the shuttle box to increase task difficulty and decrease random escape behavior and 2) adopting a new escape failure criterion for helpless behavior which is based on statistical prediction of the persistence of escape deficits.