We investigated how the updating function supports the integration process in solving arithmetic word problems. In Experiment 1, we measured reading time, that is, translation and integration times, when undergraduate and graduate students (n = 78) were asked to solve 2 types of problems: those containing only necessary information and those containing extraneous information. The results indicated that participants required more integration time to solve extraneous-information problems than necessary-information problems. However, the higher the updating, the smaller the increment of integration time. In Experiment 2, we investigated whether different problem models were provided by undergraduate and graduate students (n = 73) with different updating functions. Participants executed a lexical-decision task immediately following an integration process. The lexical-decision task comprised 3 conditions: necessary-information word, extraneous-information word, and novel word conditions. The RTs for both necessary- and extraneous-information word conditions were faster than that for the novel word condition. The facilitation amount in an extraneous-information word became weaker as the problem solver's updating function increased. These results suggest that individuals with a high updating function provide a problem model that maintains only task-relevant information, while those with less-effective updating use an approach that also considers extraneous information. These 2 experiments indicate that updating is an important contributor to the integration process and different updating abilities result in different problem models.