A production planning problem, known as the discrete lot sizing and scheduling problem with sequence-dependent changeover costs, is considered. We propose a new way of modelling the production system based on the use of a multi-attribute product structure encountered in many industrial situations. The basic idea is to describe the products as combinations of physical attributes and to exploit this description to reduce the size of the mixed-integer program to be solved. The results of our computational experiments show the practical usefulness of the proposed formulation which leads to significantly improved efficiency in the solution process.