The Social Security Disability Insurance (SSDI) system is designed to provide income security to workers in the event that health problems prevent them from working. In order to qualify for benefits, applicants must pass a medical screening that is intended to verify that the individual is truly incapable of work. Past research has shown, however, that the screening procedures used do not function without error. If screening were error-free, it has can be demonstrated that it is socially optimal to distinguish the disabled non-worker from the non-disabled, providing benefits to the disabled. In this paper we first demonstrate that if the errors in the medical screening are too large, it will not be optimal to distinguish the disabled from the non-disabled. Then, we use data on the actual quality of screening to determine first, if segmenting the non-working population is desirable, and second whether the current SSDI system relies too heavily on screening than is justified. Our preliminary conclusion is that while screening is good enough to justify some distinction in benefits, it may not be good enough to justify the size of the benefit offered.