A novel approach for performance-based design is presented, where Pareto optimality is pursued. Design requirements may contain linguistic information, which is difficult to bring into computation or make consistent their impartial estimations from case to case. Fuzzy logic and soft computing are the essential means to deal with this matter. In this work an innovative neural fuzzy system is considered for soft computing in design. The system has a neural network structure with the properties of neural tree. The nonlinear processing units at the nodes are selected as Gaussians, so that the system can be interpreted in fuzzy terms. Such a knowledge model can be subject to employment in many diverse areas. In this work it is used for a soft computing application in architectural design, where a number of linguistic information is used in the specification of requirements. The quantifications of qualitative descriptions in design are integrated into the system and fuzzy computations are carried out in a neural network framework. The application concerns a layout of multiple housing units, involving multiple, conflicting requirements, so that Pareto optimality is aimed for. This is a much desirable aid in a design process as it provides guidance for design enhancement, where the design quality underlies the guaranteed designperformance as to the specifications.