The ability to direct covert visual spatial attention to the left (LVF) and right visual field (RVF) was examined in 15 patients with mild to moderate Alzheimer's disease and 15 age- and education-matched controls using the covert orienting of visual spatial attention task (COVAT) modified to include both spatial and non-spatial cues. Subjects responded with a button press when they detected a target at a location 8 degrees to either the left or right of fixation. On 70% of trials a spatial cue was flashed at the target location before the target appeared. On 15% of trials the spatial cue was flashed at the location contralateral to where it would appear and on the remaining 15% of trials non-spatial diffuse cue preceded targets. The cue to target interval (CTI) varied between 150 and 550 ms. Mean reaction times for each cuetype in the RVF and LVF were calculated. Compared with controls, the percentage of trials excluded because of very slow reaction times was significantly greater in the Alzheimer's disease group for the 550 ms CTI. Analysis of the symmetry of reaction times to LVF and RVF targets for the 150 ms CTI enabled us to classify Alzheimer's disease subjects into three subgroups based on the hemifield of abnormally slow attentional biases. The first subgroup showed a significant slowing of reaction time to all LVF targets, the second showed a significant slowing of reaction time to all RVF targets and the third showed a significant slowing of reaction time to both LVF and RVF targets. Patients with Alzheimer's disease who showed an abnormal attentional bias performed significantly better on neuropsychological tests of memory, language and executive function than Alzheimer's disease patients with no attentional bias. Eight of the Alzheimer's disease subjects were assessed serially on at least six occasions over a 12-month period. The initial classification of abnormal attentional bias or no attentional bias was reliable for seven Alzheimer's disease subjects. One Alzheimer's disease subject, initially classified as having a slowed rightward attentional bias, in subsequent testing over the 12-month period was more consistent with symmetrical COVAT performance. Control subjects showed no attentional biases over the 12-month period and the magnitude of asymmetric attentional slowing over the 12-month period was significantly more variable in individual Alzheimer's disease subjects than in controls. The presence of subgroups of patients with Alzheimer's disease with qualitatively different COVAT performance indicates a large between-subject variability in attentional deficits in Alzheimer's disease. The presence of asymmetric attentional slowing and milder neuropsychological deficits in a subgroup of patients with Alzheimer's disease suggests that in these patients there is functional impairment of attentional areas in only one hemisphere rather than an asymmetric impairment of both hemispheres and that the neurodegenerative disease process may have been less advanced or in an earlier stage than that present in Alzheimer's disease patients with symmetric attentional performance and bilateral COVAT impairment. The preservation of asymmetric attentional slowing over time, together with the increased intra-subject variability in the magnitude of these asymmetries, suggests that asymmetrical COVAT performance represents a reliable reflection of underlying hemispheric function in Alzheimer's disease, although designation of asymmetrical attentional biases should be made on the basis of two or more sequential testing sessions.