With the proliferation of web applications, massive amounts of textual content are continuously being produced on the Internet. Machine Reading Comprehension (MRC) is attracting increasing attention for analysing and extracting information contained in text. As an important technology, MRC significantly promotes commercial value of Internet applications. Traditional MRC models adopt an extractive paradigm, which leads to unsatisfactory performance for unanswerable questions. To address this issue, we propose a cue-inspired machine reading comprehension model. Specifically, we combine sketchy and intensive reading to mimic the human reading comprehension process. The experimental result shows that our proposed model achieves better performance on several public datasets, especially when it comes to unanswerable questions.