When each edge in a network system has random multistate capacities depending on the degree of failure, the probability of successfully transmitting the required amount of information between two specified vetices is evaluated by backtracking search and decomposition algorithm. Doulliez and Jamoulle analyzed the problem using decomposition principle and the Ford-Fulkerson algorithm. However, they didn``t introduce the notion of minimal path vector, and Ford-Fulkerson algorithm cannot be efficiently adopted to the present problem. This thesis develops a systematic procedure using the concept of minimal path vectors to overcome this difficulty. Furthermore, there has been no works on the edge importance when a network has random edge capacity. This thesis adopts the importance measure used in multistate system analysis, and points out several difficulties when this measure is adopted to this problem.