In a parallel computer system, fault-tolerance is more obtainable since there are than one processor cooperated in the system. And, for solving parallel prefix computation, we need not use all processors all the time. Thus, we shall propose a fault-tolerant mechanism when we solve the prefix computation problem on faulty hypercube computers. By our mechanism, prefix computation can be down in O(log n) time on a faulty hypercube computer if there exists a fault-free subcube with size being at least a half of the hypercube, and in O(log2n) time in the worst case.