An algorithmic-evolutionary synthesis procedure is studied for generating the maximum energy recovery (MER) and minimum number of units (MNU) networks with the goal of achieving the global optimum network under pinch points. For pinched problems, sufficient conditions are proposed for determining the minimum number of units. These sufficient conditions, together with heuristic matching rules, are used to generate an initial feasible composite MNU/MER network. A split-merge network structure is introduced in order not to violate the prescribed minimum approach temperature. This initial network is successively evolved to obtain improved networks by limited heat load redistribution resulting from the pinch point. The properties and limitations of the constructions and procedures are established and the effectiveness of the heuristic procedure is illustrated with literature test problems.