With the development of communication technology and popularization of smartphones, the massive cellphone trajectory data gathered by base stations plays an important role in some applications, such as urban planning and population flow analysis. In this paper, a Movement Features-based Judging Urban Population Flow (MF-JUPF) algorithm utilizing cellphone trajectory data was proposed to deal with the issue about the population flow. First, users' activity trajectories were mined from cellphone trajectory data after data preprocessing. Second, the movement features were extracted according to the pattern of entering and leaving a city, and the parameters of these features were trained using various classification models upon real data sets. Finally, trained classification models were used to judge whether a user came in/out of the city. To enhance the efficiency and scalability, a MapReduce-based algorithm was developed to analyze massive cellphone trajectory data sets. As reported in the experimental part upon real data sets, the precision and recall of the proposed solution to judge the entering and leaving behaviors were greater than 80%. In comparison with Signal Disappears-based Judging Urban Population Flow (SD-JUPF) algorithm, the precision and recall of entering city judgment increased by 19.0% and 13.9%, and the precision and recall of leaving city judgment increased by 17.3% and 6.1%. Compared with the non-filtering algorithm, the time cost of the improved filtering algorithm was reduced by 36.1% according to the traits of these data. The theoretical analyses and experimental results illustrate the high accuracy and flexibility of MF-JUPF which has applicable values in urban planning and other fields.