Until now, an evolutionary method or empirical rule has been used to improve the initial design of batch/semicontinuous processes. In this paper, the proposed approach is to make a superstructure which fully contains all possible structures. And then suing a MINLP(mixed integer nonlinear programming) model, we obtain the optimal process network. MINLP model contains methodology deciding the candidates of mer-ging tasks and using parallel equipments. Thus, the economic savings of capital cost of batch/semicontinuous processes are possible by adding equipment in parallel for bottleneck stages or merging/splitting tasks. MINLP model is solved by using a generalized reduced gradient algorithm. The effectiveness of MINLP model in single product batch plant is demonstrated by solving example problems. Also, by inserting intermediate storage tank, we reduce the capital cost for batch/semicontinuous processes and determine the optimal size of intermediate storage tank.