This paper presents a recursive method for heat exchanger network synthesis based on a tree-decomposition strategy and a new match-selection model. The problem is decomposed as a binary tree, and each node is then categorized as either a capital-dominant or an energy-dominant problem. Subsequent decomposition of each node is determined by this dominance. The final design is obtained recursively by applying a design algorithm from child nodes to their parent node. The match-selection procedure is a hybrid method that exhibits some of the features of both evolutionary and mathematical programming methods, Matches are first allocated using a IP assignment model. This step is then followed by an evolutionary procedure in which the remaining sections of the design are treated as a new problem. The process is repeated until no saving can be discovered. The method avoids the solution of complex MINLP models, and consequently, it can be applied to solve large problems. Furthermore, it readily copes with typical constraints, such as forbidden matches and imposed matches. As a consequence, safety and layout considerations are easily incorporated into the design.