We propose two methods for exploratory image search systems using gaze data for continuous learning of the users' interests and relevance calculation. The first system uses the fixation time over the images selected by gaze in the search results pages, whereas the second one utilizes the fixation time over the clicked images and fixations over the non- selected images on the results page. A user model is trained and continuously updated from the gaze input throughout the whole session in both systems. We conducted an experiment with 24 users, each performing four search tasks using the proposed systems and compared the results to a baseline system, which does not employ any gaze data. The Gaze feedback system users viewed 22.35% more images than the users of the baseline system. A high correlation between the number of saved images and the satisfaction with the results was observed in data collected from the users of a mouse feedback system enriched by gaze data. The results show that including the gaze data into the relevance calculation in both cases increases the degree of satisfaction with the search results compared with.