A machine-administered, branch-chain automated medical history (AMH) was evaluated as a medical intake screening tool for inmates of a large metropolitan county jail. The reproducibility, validity, sensitivity and specificity of the AMH were measured and found comparable to other previously reported AMH systems. The AMH did not produce intake screening data comparable to that obtained by trained face to face medical interviewers. This, in addition to the time and personnel effort required for administering the AMH while preventing vandalization of the equipment, made the AMH impractical for use in the county jail setting. Physicians used only 21 per cent of the positive AMH responses when they were asked to formulate problem lists from the AMH data. Physicians who used the AMH in clinical settings complained that "the AMH data seem as often wrong as right." The authors observed that the number of true-positive AMH responses was matched by an almost equal number of false-positve and false-negative responses. This may account for the physician's subjective response. If so, this represents a problem to be addressed by future AMH systems. Physicians used only 55.8 per cent of the intake screening information obtained by nurse practitioners. This was also somewhat lower than expected. For this reason, medical historical information in the county jail setting might be most efficiently obtained by the provider who has clinical responsibility for prescribing treatment.