A mathematical model for synthesizing multistream heat exchanger networks is presented which eliminates the unreasonable assumption on isothermal mixing of stream branches, hence the model is more rigorous and practical. The model becomes more complex due to its large scale dimensions, nonlinear constraints and nonconvexity of objective function. Multiple local optima might also exist. It is difficult to use traditional algorithms to find the global optimization solution of this mathematical model. A Parallel Genetic Algorithm/Simulated Annealing Algorithm (PGA/SA) with the strategy of inter-subpopulation crossover and migration, generation of initial feasible population and infeasible individual repair approach is proposed to obtain the global optimization solution with larger probability. Some computational examples from literatures and an industrial project show that the proposed mathematical model and algorithm are feasible and effective.