This paper examines the problem of studying and comparing student software designs. We propose semantic categorization as a way to organize widely varying data items. We describe how this was used to organize a particular multi-national, multi-institutional dataset, and present the results of this analysis: most students are unable to effectively design software. We examine how these designs vary with different academic and demographic factors, and discuss the implications of this work on both education and education research. (Contains 4 figures, 2 tables and 2 notes.)