The ability of summarization is an important indicator of textbase understanding. Readers who are able to write an ideal summary of text can build up the macrostructure of text and extract the literal meaning from the text (Kintsch, 1998). A lot of local and global inferences were needed for macrostructure. Readers employ the three macro-rules during reading: First, readers delete both redundant and trivial information from the text. In order to do so, readers read text elements sequentially and decide the importance for each element on inclusion or deletion by their background knowledge. After deletion procedure, readers generalize related or similar concepts/sentences and construct superoridinate terms as substitutes. In this paper, two studies were presented to show how these inferences affected the construction of summary. In the first study, the purpose was to investigate the development of there types of inferences during summarization (i.e., deletion, generalization and construction). Fifty-three fourth graders, 55 fifth graders and 56 sixth graders were recruited from an elementary school in Chia-Yi City. Three versions of experimental texts, which were created to examine deletion, generalization, and construction inference-making abilities respectively, were given to participants for writing 60-words summaries. There was a significant developmental difference between fourth and sixth graders on the performance of generalization. Then, we selected low-ability participants from the 6^th grade level and matched their word-recognition and reading-comprehension scores with participants on 4^th grade. The result showed that there was a reliable difference between the low-ability group and the comprehension-match group. In the second study, the purpose was to examine how generalization inference affected children’s summarization. Sixty-five fourth graders and 53 sixth graders were recruited from an elementary school in Yunlin County. All participants were administered the Reading Comprehension growth Test, the generalization test, and the expository summarization assessment. The results showed that there was significant developmental difference on the performance of generalization inference. In addition, the generalization inference reliably predicted the performance of summarization for the fourth graders.