Abstract. Visibility query is fundamental to many analysis and decisionmaking tasks in virtual environments. Visibility computation is time complex and the complexity escalates in large and dynamic environments, where the visibility set (i.e., the set of visible objects) of any viewpoint is probe to change at any time. However, exact visibility query is rarely necessary. Besides, it is inefficient, if not infeasible, to obtain the exact result in a dynamic environment. In this paper, we formally define an Approximate Visibility Query (AVQ) as follows: given a viewpoint v, a distance and a probability p, the answer to an AVQ for the viewpoint v is an approximate visibility set such that its difference with the exact visibility set is guaranteed to be less than with confidence p. We propose an approach to correctly and efficiently answer AVQ in large and dynamic environments. Our extensive experiments verified the efficiency of our approach.