Spatial neglect has for decades been a phenomenon of interest for perceptual and neuropsychological researchers. We provide a Bayesian framework for understanding characteristic neglect patterns in line bisection tasks. This involves re-conceptualizing neglect behavior as a consequence of rational perceptual inference when the brain receives less precise information from one side of space. A Bayesian observer who perceives the left and right endpoints of a line with independent degrees of (im)precision could have leveraged their top-down expectations to compensate for having asymmetric uncertainty about the stimulus. Our Bayesian model provides a basis for modeling both clinical and healthy response patterns and de-confounding individuals' stable attentional deficit from their more-changeable and context-dependent expectations. The model accounts for the paradoxical cross-over effect in line bisection by interpreting it as a natural consequence of the strong influence of prior expectations, particularly under conditions of reduced attention to one side and the resulting imprecision. The fit parameters from the Bayesian model has strong correlations with other spatial neglect tests and are able to indicate neglect patients from the controls. This novel perspective enriches our understanding of spatial neglect as well as top-down influences on attention-relevant perception more broadly, and has the potential to inform future neural-based studies and clinical decision making.