Current methods for analyzing eye movement data focus on spatial information of eye movements,such as number of fixations on a set of predefined regions of interest (ROIs) or distribution of fixation locations on a heat map.In contrast,temporal information such as transitions between different ROIs is typically overlooked.Also,using predefined ROIs does not take individual differences into account and may be subject to experimenter biases.Here we propose a hidden Markov model (HMM) based approach to analyze eye movement data,which takes individual differences in temporal and spatial information of eye movements into account.We use a variational Bayesian framework for Gaussian mixture models to estimate fixation distribution of each individual to obtain personalized ROIs.We then model each individuals fixation location and transition data using an HMM.The HMM summarizes the individuals eye movement pattern with both fixation distribution/ROIs and transition probabilities among the ROIs.Here we apply this method to analyze eye movements in face recognition in an old/new judgment task.We show that:(1) by clustering Asian participants HMMs according to their similarities,holistic and analytic eye movement strategies emerged naturally in a data-driven fashion.The analytic group generally had longer response times and made more fixations,although the two groups did not differ significantly in recognition performance.(2) Similar eye movement strategies could be found in both Asians and Caucasians;however,the percentages of people using the strategies differed between the two ethnic groups.(3) Participants correct and wrong recognitions were associated with distinctive eye movement patterns,and the difference lied mainly in the transition probabilities among regions instead of the fixation locations alone.These discoveries are not possible with traditional analysis methods that do not take individual differences and temporal information into account.