Generalizing design knowledge is a process of extracting knowledge fro
m design-provided data. Although it is one of the most common ways of
''learning'' design knowledge, generalization has a fundamental weakne
ss: except for special occasions, the results of generalization can ne
ver be validated. Inquiries into generalization have therefore dealt w
ith questions of what are the best criteria for guiding the generaliza
tion. This paper argues that generalizing design knowledge involves th
ree essential tasks; knowledge representation, a description language
for design examples, and generalization operators. An application of g
eneralizing empirical networks from design examples is described to il
lustrate these three tasks in the development of a generalization syst
em.