A problem which often arises in analyzing survey data concerns finding a way to partition a large number of variables into a more parsimonious set. The procedure most often used for this partitioning is linear factor analysis (Harman 1960; Morrison 1967). Other means of grouping variables are considered here; specifically, a nonlinear measure of association is substituted for the simple correlation coefficient. Alternatives to the conventional factor analysis approach for grouping variables are suggested. These suggested approaches are then applied to a common data set involving a new product introduction.