Agro-product safety incidents can be disseminated quickly on the Internet, which are easy to cause great social impact or lead to network public opinion events. Therefore, it is urgent to identify agricultural safety crisis events online. An online detection method of agro-product safety incident based on improved Single-Pass is proposed. The method of computing the term weight is improved by dividing the text into several blocks and dynamically updating the term frequency; the similarity measure method of Single-Pass algorithm is improved by introducing time distance. Compared with the traditional method, the false alarms and misses of the system are obviously reduced. The proposed method can effectively detect agro-product safety incidents online, which can be used for dynamic monitoring of agro-product safety crisis events on the internet.