The underlying premise of this paper is a comparative study of a specialized corpus (Chemical Engineering Corpus) and general corpora (BNC, BNC baby). Chemical Engineering Corpus is divided into two sub-corpora, Chemical Engineering Textbook Corpus and Chemical Engineering Article Corpus. Two sub-corpora consist of 5 million words compiled from 31 textbooks and articles of 5 journals, respectively. This paper is to explore the keywords and the structural features of 4-word bundles in two sub-corpora using WordSmith 6.0 and offer authentic language sources contributing to effective language teaching and learning. The analysis shows that two sub-corpora have genre-specific features remarkably distinct from general corpora. Therefore, the data used in the analysis can be used to help language instructors provide relevant language teaching materials, and make learners improve their reading and writing skills and have a better understanding of textbooks and articles they will study in their courses.