A recent survey reported that, in order to identify accident blackspots, most highway authorities use annual accident total (AAT), without making any allowance for any measure of exposure or the nature of the site. An alternative, and apparently improved criterion, known as potential accident reduction (PAR), has been proposed. The basis for this is that statistically significant relationships can be established between accident frequency and traffic flow, for a variety of site categories. PAR is calculated, for a site, as the difference between the observed number of accidents and the expected number for that type of site with that level of flow. If this expected frequency can be estimated accurately, then PAR should inevitably perform better than AAT, since AAT tends to identify high flow sites which do not necessarily have the potential for accident reduction. This work attempts to quantify the additional benefits to be gained by the use of PAR instead of AAT. It demonstrates that, because of the inaccuracy of the estimation of the expected frequency at a site required in PAR, it is quite possible for AAT to perform as well as, or better than PAR. If the sites, identified as blackspots, are then given some remedial treatment, the analysis of the effectiveness of this treatment must allow for what has become known as the "regression-to-mean" effect. A method which provides a proper basis for such before and after studies is described.