Brain damage is often accompanied by homonymous hemianopia, but few therapeutic approaches exist for visual field deficits. In this open pilot study we describe a computerized training program which may possibly reduce the size of the 'blind' visual field in patients with homonymous visual field deficits. Various stimuli to test light perception and discrimination of colors and shapes were presented on a monitor which permitted the examination or training of the central section of the visual field up to about 25° vertical and 40° horizontal eccentricity. Eleven patients trained at home for 1 h each day for a total of 80-300 h. Their results were compared with those of three patients who opted not to participate in the training procedure or those with very little therapy. These latter subjects had a slight decrease in the visual field size after about 1 year. In contrast, the treatment group displayed a reliable enlargement of visual field size. This was revealed by a significant improvement in the detection of small light stimuli, an increase in the ability to discriminate colors and a minor, but notable, improvement of shape discrimination in the blind areas of the visual field. Additional training of shape recognition led to further improvement of shape discriminations, even when the patients trained with very different kinds of shapes, e.g. lines or letters. Outcome depended on age of the patients and the size of the lesion, but it was independent of on-set of treatment and cause of the lesion. Only two of the 11 patients with treatment showed no significant improvement. This study suggests that regular home training of the 'blind' visual field with computer-controlled stimuli may lead to improvement in vision. However, because of the following methodological limitations results are only preliminary: (1) the trial did not contain a true placebo group, (2) the patients were not assigned randomly to a control or treatment condition, (3) the lack of defined inclusion criteria considerably increased the variance in neuropsychological performance, (4) because the experimental design was not double blind, experimenter bias cannot be ruled out, and (5) the conditions of the home training could not be standardized. The results warrant a larger randomized, double-blind controlled trial.