A multiyear study was undertaken to examine the effects of the use of a computer algebra system (Mathematica) on student learning in multivariable calculus at the University of Oklahoma. The research was both quantitative and qualitative, including pre- and post-questionnaires administered to 585 students over four semesters as well as task-based interviews. Although in early iterations of the data collection there were some apparent effects on spatial visualization skills, with the full data set it became apparent that these differences were artifacts of student self-selection into the "experimental" and "control" sections. In particular, gender and major course of study were responsible for considerable differences in visualization performance.