This study aims to analyze the writing characteristics of middle school students through semantic network analysis. The study results are as follows. First, the number of words (nouns) used in the text was higher in the high school group than in the middle school group, and in the case of the number of frequency-reflecting nouns indicating lexical diversity, the high school group had more words than the middle school group, but less than the upper-level group. Thus, students with higher writing levels have the ability to write longer texts with fewer words. Second, as a result of checking the structure of the semantic network, the overall degree of connection between words was higher in the high school group than in the middle school group, but by writing level, the degree centrality and density were lower than those of the upper and lower levels. According to the modularity index, which is a clustering method based on the difference in density, it was confirmed that the degree of semantic differentiation of texts increased as the school and writing performance levels increased. Third, as a result of analyzing the pattern of the semantic network graph, the higher the school and writing levels, the denser the connection from the center to the periphery. Conversely, it was confirmed that the semantic network pattern appeared with a large difference in density between the core and the periphery and weak overall cohesiveness of the network in writing with a low writing level. Fourth, as a result of analyzing the word centrality of individual texts, the centrality of words used to reveal the author’s standpoint or develop logic, as well as content words related to the writing topic, were high in the upper-level texts, unlike the lower-level texts. This can be seen to be related to the aspect of knowledge-transforming writing, in which the more proficient the writer, the more his knowledge is transformed to induce the reader to accept the content of the text.