Traffic accidents are responsible for about 3,000 deaths and $25 billion in economic losses annually in Canada. One way for transportation authorities to improve safety is to identify potentially hazardous roadway elements through network screening. The process of network screening is a low-cost statistical analysis of highway safety data, which yields a ranked list of sites to be investigated in detail. Critical issues of two network screening methods are investigated in this thesis. The first method is a peak-searching algorithm for screening roadway segments, with attention focused on threshold values of a key user-selected variable, namely the coefficient of variation. The second method examined is a method of screening for high proportions of specific accident types. For this method, parameter estimation techniques are compared, and the effect of the 'critical proportion,' a key user-selected variable in the method, on site rankings is investigated. In addition to the two network screening methods, an investigation is carried out into some aspects of safety performance function calibrated using negative binomial regression. Specific attention is given to how the negative binomial dispension parameter changes over the range of some independent variables.