A large number of tragic drug-induced events worldwide have warned the necessity to pay attention to the safety of drugs. Drug regulatory authorities and marketing license holders are obliged to collect information on adverse drug reactions as early as possible and develop different risk management plans for different drugs to maximize the safety of drug users. In the whole life cycle management of drugs, signal detection is an important part of post-marketing safety monitoring. This paper introduces the basic principle of disproportionality analysis used in post-marketing adverse drug reaction surveillance and potential biases and bias control methods in data analysis, and discusses machine learning method for signal detection and the standardization of signal detection report, so as to improve the accuracy and timeliness of signal detection.