Patients with relatively mild unilateral spatial neglect underwent the line bisection task with self-estimation, where they were asked two different questions about the relationship of their subjective midpoint to the line. For the first trial, all patients responded by "yes" when asked whether their subjective midpoint was placed at the center of the line, but by "left" or "right" when asked to indicate the longer segment of the bisected line. In the subsequent trials, they repeatedly reported that their subjective midpoints were placed at the true center, even after they had recognized their bisection errors when indicating the longer segment. Of the 32 patients, 28 kept on showing this pattern of responses in almost all the 30 trials. Moreover, the patients who could estimate the longer segment showed no improvement of line bisection throughout the trials. Their performances strongly depended upon the task demand regarding the subjective midpoint. These results demonstrate that the patients treated the two related tasks as independent. Accordingly, they could not improve their performances on one task with the feedback from their performances on the other. The lack of association between the responses to two related task demands may be one of the factors that prevent patients with unilateral spatial neglect from improving the deficit.