Epilepsy refers to a set of chronic neurological syndromes characterized by transient and unexpected electrical disturbances of the brain. There are a large number of patients suffering from epilepsy in the world. Epilepsy cannot be cured eventually, but 70% of seizures can be kept under control by drugs. Electronic health records (EHRs) of epileptics contain a wealth of information for personalized medicine prescription, providing a large number of data resources. Based on real medical electronic cases for large data analysis, this paper proposes a drug recommendation system based on implicit feedback and crossing recommendation (IFCR) to help doctorschoose right drugs. The proposed system aims to analyze the patients’ medical history and similar patients’ in order to find the relationships between syndromes and drugs. Comparing our system with the one based on artificial neural network (ANN), the proposed algorithm performs much better than ANN in terms of the recall rate with a 30% improvement. However, two algorithms have different performance on the precision rate. In general, the performance of IFCR is better than that of ANN. Finally, we analyze the recommendation results of two algorithms and discover it is possible to propose an ensemble model to compile IFCR with ANN.