Abstract: We present initial results from a project, in which we examined feasibility of inferring web page relevance from eye-tracking data. We conduced a controlled, lab-based Web search experiment, in which participants conducted assigned information search tasks on Wikipedia. We performed analyses of variance as well as employed classification algorithms in order to predict user perceived Web page relevance. Our findings demonstrate that it is feasible to infer document relevance from eye-tracking data on Web pages. The results indicate that eye fixation duration, pupil size and the probability of continuing reading are good predictors of Web page relevance. Our work extends results from previous studies of text document search that were conducted under more constrained human-information interaction.