Item selection is one of the most important indicators of the effectiveness of learning control. Early researchers focused on the influence of item difficulties on item selection. It is believed that learners will monitor the item difficulties, and choose to study the easer items first, especially when total study time is limited. Recently, researchers found that when facing materials with different values, learners choose to give priority to study the items with high value. However, when they came to the conclusion that learners will choose items according to their difficulties or values, they did not control the other factors strictly. In the present study, we want to know when facing items with different difficulties and different values, what kind of items will learners choose to learn, the easiest one, the one with the highest value, or the one with the highest expected return? And how does the selection process take place? In experiment 1, 3 items (each item contains a Spanish word and its corresponding Chinese translation) were presented on a 21 inch screen. One was easiest (highest pass rate), one had the highest point, and the other one had the highest expected return (pass rate × point). Pass rate and point of the items were marked separately above and below the items. Participants were asked to choose one of the three items for the subsequent learning phase. In experiment 2, 2 items with different pass rate and different point were presented, and participants were asked to choose one from the two items for the subsequent learning phase. Eye movement data was recorded using an EyeLink 1000 Plus eye tracker sampling at 1000 Hz. All words were presented in white SimSun 21 point font, on a black background. If participants selected items by calculating the expected return of the items, there would be more saccades between different attributes of the same item than saccades between same attributes of the different items. The main results were as follows: (1) In Experiment 1, there was a significant main effect of item type, F(1, 29) = 225.41, p < .001, ηp2 = .89. Participants chose more items with the highest expected return than other items (ps < .001).(2) In Experiment 2, there was no significant difference in first fixation index, proportions of dwell time, proportions of fixations and average pupil diameters between pass rate and point, ts ≤ 1.54, ps > .05. (3) In Experiment 2, there were more saccades between same attributes of the different items than saccades between different attributes of the same item t(25) = 2.29，p < .05，Cohen’s d = .46. The findings suggest that when learners facing materials with different values and different difficulties, their item selection is not only based on difficulty or value, but based on comprehensive consideration of difficulty and value. And the comprehensive consideration is not by performing an expectation computation, but by comparing the same attributes of different items. The results are compatible with uncompensated models.