Readers should construct a coherent discourse during reading comprehension. The ability to build coherence has been examined using coherence and cohesion judgment tasks. Although eye-tracking studies have been conducted on building coherence or processing cohesion among native language users, few such studies have been conducted with second language learners. This study investigated the process of conducting a task using eye-tracking with 30 Japanese university students who were asked to judge the coherence between two sentences on a screen. The 24 stimulus items consisted of four conditions (with or without coherence and cohesion), and six items were assigned to each condition. The fixation duration in some areas and frequencies of the regressions were measured. The results showed that the difference in fixation duration between conditions was not statistically significant, but the frequencies of regressions were significantly different, with incoherent-cohesive pairs having more regressions than incoherent-incohesive pairs. Thus, the study revealed that readers might make more inferences because of the contradictory information of the condition, which partly supports previous research on native participants.