This paper presents the idea behind resume screening system used by human resources. The implementation of NER offers a practical scenario for natural language processing in the real world. We first review the traditional solutions to sequential labeling tasks but then point out their drawbacks. Non-linear neural networks including LSTM and its variants are then introduced. By experimentally investigating the performance of four models on different NER tasks, we conclude that BLSTM-CRF with character-level embedding obtains the best performance on all evaluation matrices. Finally, the case study is employed to analyze the model performance, and offer a thorough bridge connecting the confusion matrix and the experimental data set.