A novel computational methodology for architectural design will be developed that is based on interaction between architect and multi-objective evolutionary computation. Existing interactive multi-objective optimization approaches require that all objective functions are mathematically formulated prior to the interactive process. Therefore the approaches found applications mainly in the engineering domain up till now. The originality of the proposal is that the qualitative preferences of an individual architect are collected during multiple interactions between the human and ongoing computational search process, so that his qualitative judgment is brought into the form of computational model, yielding an additional objective function. For this the method fuzzy neural tree is used. It is an innovative neuro-fuzzy system based on Gaussian information processors, where the Gaussians can be interpreted as fuzzy membership functions or likelihood distributions at the same time. Due to the computational judgment model, the search for best solutions need not terminate at the moment when fatigue diminishes the precision of the architect’s qualitative judgments it can be carried on until solutions are found that maximally satisfy both, the quantitative and qualitative criteria. Next to enhanced design effectiveness, from the theoretical viewpoint the study is to contribute to our understanding, how human establishes abstractions from a low number of trials in the formation of his cognition during design. This may provide new clues on effective knowledge conveyance in architecture education.