In order to the effective application of the massive detection data in food safety early warning, this paper analyzed the characteristics of the food detection data, and the insufficient of traditional Apriori algorithm on food detection data, then proposed the filtering algorithm, which is a pre-components of Apriori algorithm. An early warning model was established, which was applied to excavate the real oil detection data, and the potential safety problems were founded to make an early warning. Compared with the Apriori algorithm, the improved algorithm abandoned a lot of pseudo-association rules, and also could effectively enhance the efficiency and accuracy of food safety early warning, which has a very important practical significance.