Although control variables are the reason for multiple regression, surprising little attention is given to the process of locating and selecting the controls. If important controls are omitted, estimates can be biased and inconsistent whereas using too many controls can reduce efficiency and mask effects. We advocate a much broader search for controls than is customary, at least in the crime literature, and then the use of the general-to-specific methodology to select the relevant controls. We illustrate this procedure using a fixed effects panel data design evaluating the effect of legalized abortion on crime.