After a review of the classical construction of regular fractional factorial designs that is treated in every textbook on experimental design, I will concentrate on developments in the past decade. A fundamental issue is the selection of an "optimal" design, and, in the case of blocked experiments, an "optimal" blocking scheme as well. Maximum resolution, minimum aberration, maximum estimation capacity and other criteria for choosing regular fractional factorial designs will be discussed in details. Then I will move on to discuss non-regular designs and supersaturated designs, including a generalization of the minimum aberration criterion. Unlike regular fractional factorial designs, non-regular designs have complex alias structures and interesting low-dimensional projection properties that have important practical implications. I will try to present a unified theory.