The ’block’ concept has been applied successfully to heat exchanger network (HEN) design problems(1,2). The methodology first produces simple initial designs by employing simple heuristic rules; subsequently these initial designs are optimised using NLP techniques. In this paper, an extension of the block concept to mathematical programming is discussed. The block concept is based on physical insights from the composite curves. The HEN synthesis problem is decomposed into a number of blocks determined such that streams in each block have similar characteristics. When applied to mathematical programming, the block concept leads to the block-based superstructure which is similar to the superstructure suggested by Linnhoff and his co-workers and Yee and Grossmann(3). The difference is that this work is based on blocks (large enthalpy intervals) which are derived from physical insights(1). As a consequence, the overall superstructure for a problem is much simplified and thus the dimensions of the mathematical models are greatly reduced. Moreover, since the block-based superstructure is based on the composite curves, good initial solutions can be derived which also assist the subsequent optimization process. Combining the block concept with non-linear programming techniques therefore allows automatic generation of optimal or near-optimal networks.