One of the major issues related to the conceptual phase of 3D scene synthesis is the acquisition and incorporation of qualitative aspects into the design process. Our motivation is to alleviate the task of synthesizing a scene conforming to the particularities of certain morphological or other intuitive demands. We address the feasibility of such a framework, through the study and integration of methods for the acquisition of qualitative knowledge as well as for the guided generation of scenes. The proposed framework comprises a knowledge obtainment stage, where the machine learning mechanism acquires the qualitative implications of a series of examples and a scene generation stage, where the evolutionary mechanism conducts a partial exploration of the solution space of scenes, compliant with a given set of quantitative requirements as well as to the acquired qualitative knowledge. The paradigm domain is that of architectural design with the aid of declarative modeling by hierarchical decomposition.