Patients with left unilateral spatial neglect following right hemisphere lesions usually err rightward when bisecting a horizontal line. For very short lines (e.g. 25 mm), however, leftward errors or seemingly 'right' neglect is often observed. To explain this paradox of crossover in the direction of errors, rather complicated models have been introduced as to the distribution of attention. Neglect may be hypothesized to occur in representational process of a line or estimation of the midpoint on the formed image, or both. We devised a line image task using a computer display with a touch panel and approached the representational image of a line to be bisected. Three patients with typical left neglect were presented with a line and forced to see its whole extent with cueing to the left endpoint. After disappearance of the line, they pointed to the right endpoint, the left endpoint, or the subjective midpoint according to their representational image. The line image between the reproduced right and left endpoints was appropriately formed for the 200 mm lines. However, the images for the shorter 25 and 100 mm lines were longer than the physical lengths with overextension to the left side. These results proved the context effect that short lines may be perceived longer when they are presented in combination with longer lines. One of our patients had an extensive lesion that involved the frontal, temporal, and parietal lobes, and the other two had a lesion restricted to the posterior right hemisphere. The image for a fully perceived line may be represented far enough into left space even when left neglect occurs after a lesion that involves the right parietal lobe. The patients with neglect placed the subjective midpoint rightward from the centre of the stimulus line for the 100 and 200 mm lines and leftward for the 25 mm lines. This crossover of bisection errors disappeared when the displacement of the subjective midpoint was measured from the centre of the representational line image. Left neglect may occur consistently in estimation of the subjective midpoint on the representational image, which may be explained by a simple rightward bias of attentional distribution.