Urban spatial cognition is a classic research topic of behavioral geography. In this study, we selected the main urban districts of Beijing and investigated the spatial cognition of graduate students of the University of Chinese Academy of Sciences using a combination of methods of cognitive map and questionnaire survey. We conducted the surveys in three stages of the same students and asked them to draw a sketch of the city of Beijing, then classified the sketch maps and counted the identified ground features. Based on statistical data, we analyzed the development process and the characteristics of differences of sketch maps of the surveyed students. The research shows that different from Kevin Lynch's five elements of space imagery, the basic elements of cognitive maps are three types: paths, domains, and landmarks, in which landmarks occupy the absolute dominant position. According to the combination of identified ground features in the sketch maps, the cognitive maps can be divided into three types: scattered structure, topological structure, and transitional structure between them, with scattered structure maps constituting the absolute majority. With the progression of time, imaginary space of the city gradually became complicated, and the spatial structure changed from single core to multiple cores; urban spatial cognition of the survey respondents showed the characteristics of personal preference, spatial non-equilibrium, individual difference, and incompleteness of the cognitive process. Beijing urban spatial cognition of the surveyed students is limited mainly because of the long distance from the target places, huge spatial scope of the city, short duration of stay of the students, and the respondents' own attributes that increase the difficulty for the students to imagine ground features of Beijing.