Finding patients who meet certain medical criteria in a clinical trial for qualification is an important aspect of reliable medical research. However, this can be challenging because of complexity of medical research and the difficulty of translating medical criteria into a database query. An efficient alternative approach would be to examine the clinical narratives in the patient’s medical history. In this study, we proposed an automated multi-criteria classification model for identification of clinical criteria met by candidate patients for a clinical trial. The model leverages Convolutional Neural Network (CNN) and sentiment embedding features in solving the problem of identification criteria met by a patient’s medical narrative. We adopted an encoding approach for extraction of features. Our results show significant improvement in the use of deep-learning and feature embedding for training. The original result submitted to Track 1 of n2c2 used one-level encoding features with a micro F1 score of 0.7526. In the two-two-levels encoding model, the distance between input and encoding features has reduced. This significantly improved the classification rating of our model with a micro F1 score of 0.8718. The outcome of the study have shown that Automating identification of patients for clinical trials using natural language processing and machine learning techniques saves the time required to recruit patients as well as the benefit of moving unwanted bias.