We identified all incident cases of clinically diagnosed Alzheimer's disease among the population of Rochester, MN, with onset from 1965 through 1974. A control (selected from the same community) was matched to each case by age, race, sex, and length of stay in Rochester. We studied 274 case-control pairs. Using the records linkage system available for residents of this community, we abstracted information on the occurrence of head trauma with loss of consciousness (LOC) from the medical records of both cases and controls. There were 5 pairs in which the case suffered an episode of head trauma with LOC but the control did not, and 4 pairs in which the control suffered an episode of head trauma with LOC but the case did not. We failed to detect a statistically significant difference between the 2 groups. This study overcomes many problems encountered in previous case-control studies reporting this association. The sample size was large; severity of head injury was documented in high-quality medical records; data about head injury were recorded before the onset of dementia: equal quality of data were available for cases and controls; and, since the study was population-based, there was no selection bias for cases or controls.