We have analysed the statistical characteristics of streak artefacts on CT images using the statistics of extremes, and have devised a new method of evaluating streak artefacts on CT images. The CT images of four polymer tubes placed on the chest wall of a commercially available chest phantom were used as the target objects for our analysis. 40 parallel line segments with a length of 20 pixels were placed perpendicular to numerous streak artefacts on the polymer tube image, and the largest difference between adjacent CT values in each of the 40 CT value profiles of these line-segments was employed as a feature variable of a streak artefact; these feature variables have been analysed by extreme value theory. Using the mean rank method, a Gumbel distribution was shown to be the most suitable extreme value distribution for the largest difference between adjacent CT values in each CT value profile. This enabled us to demonstrate that the streak artefacts on CT images can be statistically modelled by a Gumbel distribution. Both the location parameter and the scale parameter of the estimated Gumbel probability density distribution were large on the CT slices in which the shoulder bone or liver was included.