Abstract: Mathematical programming approach has been widely used for the systematic synthesis of process flowsheets. In this paper, a new logic and heuristic knowledge based MINLP approach to process synthesis is presented, which makes modeling more easily and systematically. In a typical superstructure, there are three types of unit: equipment, inlet/outlet stream and mixer/splitter. Each equipment is associated with a binary variable y i ( i =1,2,..., N ), and each inlet/outlet stream is also associated with a binary variable z i ( i =1,2,..., M ). The topology of the superstructure can be divided into 4 connection types as follows: 1-1 connection relations, 1- n connection relations, exclusive 1- n connection relations and m-n connection relations. With the logical representations of each relation, the superstructure can be converted to a logic-based MINLP model, which will be solved directly or through its algebraic forms. Operating experiences are expressed in logical forms, so that they can be integrated into the MINLP model as hard-logic or heuristic knowledge. This approach was used in a sample of utility system synthesis, which supplied three types of steam, several types of mechanical energy, electricity and water for a petroleum refinery. The synthesis result was compared with the hierarchical decomposition approach and the traditional mathematical programming approach, which showed that this approach could get better results than hierarchical decomposition. In comparison with the mathematical programming approach, this technique reduced the number of variables and the solving space, so the model size and computing time were decreased.