Despite forty years of revolution in the tools available for statistical analysis, the current academic tradition in statistics is remarkably similar to the pre-computer tradition. This tradition is rooted in parametric modeling, least squares, and linear models. The paper argues for a shift in emphasis away from parametric modeling and data reduction to graphical display, from omnibus optimal techniques to those that are more context-specific, and from goals of objectivity to goals of revelation. It is suggested that certain modern topics should be included in a basic statistical education, and that more academic work is needed to achieve this goal.