Along which paths should experimental statistics develop? By recognizing the selection of an “experimental design” as the selection of a pattern, as only a small part of designing an experiment. By developing further the possibilities of restricted randomization. By learning to consciously balance bias and variability, especially in regression situations. By developing experimental patterns, such as pieces of mixed factorials, which provide desired properties at far lower cost and by using estimates of only 90–95% efficiency. By recognizing the effect of distinct aims in diversifying well-chosen methods. By looking to new sources of stimulation, such as experiments involving very many factors, problems of tolerance design and application, and those statistical techniques really appropriate to research. By looking more and more frequently at broader canvas, considering investigations rather than single experiments. To deal effectively with this broader canvas, statisticians must consider indications as well as conclusions (keeping the two concepts separate), must work hard on problems of mutual understanding, must seek new sorts of real problems, must reshape old tools to new ends, must have a greatly increased concern with problems of choosing the structure within which the analysis is performed, and must learn to use the null hypothesis really appropriate to the situation. They will also need to step aside and consider other broad fields, where such words as experimentation and investigation are not only inappropriate but dangerous. The development of “evolutionary operation” as an operating tool in production is but one instance of what is possible.