This paper presents a freeway incident detection model that was developed using fractal dimension analysis of speed and occupancy data. The success of fractal dimension analysis in modelling complex behaviour of real-world systems in fields as diverse as seismology, meteorology and economics provided the motivation for investigating this technique in the development of incident detection models. The algorithms developed in this study were based on the notion that traffic parameters upstream of incidents and bottlenecks show substantial irregular behaviour when compared with downstream conditions. Fractal dimension analysis was used to provide a measure of the irregularity in traffic parameters. A number of fractal models based on a combination of smoothing and recursion tests on speed and occupancy data were developed in this study. These models were developed and tested using a real-world data set of 100 incidents. The best performing model was identified as that implementing smoothed fractal speed and occupancy inputs based on data collected from dualloop detectors embedded in the pavement of the freeway. The results demonstrate the feasibility of using fractal dimension analysis for incident detection and confirm that superior incident detection performance is obtained when speed data is available for testing. An evaluation of the model's performance against a number of other models also show that smoothed fractal models outperform a Comparative (California) incident detection model that was developed using the same data set. The model is, however, outperformed by a Neural Network model. The results reported in this paper provide a comprehensive evaluation of a number of automated incident detection models currently used in Australia and provide directions for using fractal analysis to further enhance their performance. Language: en