The effect of network configuration on the optimal cleaning schedule and total cost of a membrane desalination plant has been investigated. The system is analysed using a discrete time interval approach, assuming an exponential decay in membrane permeability and perfect regeneration at cleaning. The scheduling problem was solved as a mixed-integer non-linear programming (MINLP) problem using both (i) fixed network configurations and (ii) a flexible network superstructure. In the fixed network analysis, both the number of modules per unit in each stage and the optimal cleaning schedules were determined simultaneously. An initial investigation of linear networks found that a two-stage network was more attractive than single and three stage networks. Eleven two-stage network configurations were subsequently studied: 1:1, 2:2 and 2:1 networks gave the best performance. Further investigation indicated that a tapered-flow 2:1 network yielded a marginally better total cost. The scheduling and cost results were very sensitive to the form of the cost model used. In the flexible superstructure approach, the network configuration, number of units and operating schedule were all determined simultaneously. This approach identified the 1:1 configuration as the optimal design but the total cost calculated was 1.4% higher than that obtained in the fixed network analysis. This stimultaneous superstructure approach was found to consistently converge on networks with larger numbers of modules;md hence larger total costs.