Opposite-direction collision (ODC) accidents are more harmful than other collision types and often cause severe casualties and property losses. Based on the ordered discrete choice model, this study selects the data of 19623 ODCs from January 2010 to June 2018 in Victoria, Australia. The ordered logit model, the random-parameters ordered logit model and the heteroscedastic ordered logit (HOL) model are used to quantify the driver injury severities in ODC accidents. The study shows that HOL model can exploit more potential injury severity information of drivers involved from the original ODC accident data properly. The model results showed that some variables (e.g. elderly drivers, female, high speed limits of more than 50 km/h, number of injuries) increases the possibility of severe and fatal injury outcomes in ODC accidents significantly. In contrast, some variables, including give way sign control, non-darkness conditions (i.e. daytime, dusk/dawn, night lighting), slippery pavement, non-frontal collision types, number of individuals involved, and types of vehicles involved (small truck, bus/passenger car, large trucks/heavy vehicles, other), significantly decrease the serious injury and fatality probabilities in such accidents. Based on the model's findings, countermeasures are recommended, such as ongoing safe driving education, enforcing stricter speed limit rules, and setting up warning signs and lighting facilities on dangerous locations and segments.