After decades of work in pattern recognition, humans are still considered the best recognizers of images and symbols especially in unconstrained everyday applications. This has made the human visual model a major topic of interest in pattern recognition research. A number of studies have presented promising recognition models that incorporate different aspects of the human model such as selective attention, biologically plausible saliency detection and top-down recognition. On the other hand, the last hundred years of research in human eye movement behaviour has revived the ancient philosophical idea that we see in our mind’s eye. Several computational models of eye movement control were suggested that successfully predict eye movement behaviour demonstrating a close coupling between eye movements and underlying oculomotor and cognitive processes. In the present study, the author evaluates a combined approach to identifying features of interest for Pattern Recognition applications. In the data collection stage, sixty participants are asked to verbally identify fifty-four problematic and twenty prototypical handwritten digits. Both verbal responses and visual fixations are recorded for further analysis. In the analysis stage, a smaller set of ambiguous digit images is identified based on how often participants change their minds about the numeral they represent. For each digit, visual fixations are grouped based on the numeral that participants called out. Each fixation group is then combined into a single fixation heat map. Results show that by comparing and contrasting heat maps for a given digit the features deemed most disambiguating by the human model can be identified.