Hidden Markov models (HMM) can describe the spatial and temporal characteristics of eye-tracking recordings in cognitive tasks. Here, we introduce a new HMM approach. We developed HMMs based on fixation locations and we also used image information as an input feature. We demonstrate the benefits of the newly proposed model in a face recognition study wherein an HMM was developed for every subject. Discovery of regions of interest on facial stimuli is improved compared to earlier approaches. Moreover, clustering of the newly developed HMMs lead to very distinct groups. The newly developed approach also allows reconstructing image information at fixation.