Print unevenness is one of the most important defects in modern printing technologies. Understanding the complete phenomenon of such low-contrast inhomogeneity requires reliable analysis and modelling based on an experimental evaluation of visual perception. The evaluation however is not viable due to the limited available characteristics of physical test prints and their reproduction methods. In visual quality assessment problems, sample sets are limited to a narrow subset from an unknown distribution. Therefore, it is necessary to generate an artificial sample set covering a wider range of phenomenon instances with a possibility to control the sampling distribution. In this study a process of generating random stimuli for the psychometric experiment of print unevenness evaluation is proposed and based on the process a refined perception model parameter analysis is performed. We extend Thurstone's law of comparative judgement to the continuous case, where the stimuli are not limited to a certain number, but generated according to a predefined distribution. The psychometric experiment results are further utilized in perception model parameter analysis, and consequently, a refined model is devised. The selected approach is suitable to cases where the range of the initial stimuli set is not wide enough for a reliable perception model estimate.