—We present a framework to compute the visual hull of a polyhedral scene, in which the vertices of the polyhedra are given with some imprecision. Two kinds of visual event surfaces, namely VE and EEE surfaces are modelled under the geometric framework to derive their counterpart object, namely partial VE and partial EEE surfaces, which contain the exact information of all possible visual event surfaces given the imprecision in the input. Correspondingly, a new definition of visual number is proposed to label the cells of Euclidean space partitioned by partial VE and partial EEE surfaces. The overall algorithm maintains the same computational complexity as the classical method and generates a partial visual hull which converges to the classical visual hull as the input converges to an exact value. Keywords-visual hull; shape from silhouettes; imprecise input; solid domain; quadratic surface