Divided differences are the most natural way to verify derivative values computed by Adifor. Unfortunately, the accuracy of a divided difference depends heavily on the stepsize (or stepsizes). Consequently, to reliably verify Adifor derivatives, we seek a method for selecting stepsizes that minimize the error. This technical report gives a method for selecting a stepsize that will give the 'most accurate' finite difference approximation. The mathematics that justifies the selction method is based on the work of Gill, Murray, and Wright.