Serious and similar chemical accidents frequently occur, which prompts continuous exploration of an effective way to prevent chemical accidents. The purpose of learning from accidents is to prevent the recurrence of similar incidents. An in-depth exploration of the causes is a prerequisite to ensure the effectiveness of learning from accidents. The interactions among causes can explain how a particular sequence of events results in an incident with unpleasant consequences. This research aimed to provide a more concrete and intuitive understanding of the interaction of causes and represent the routes to failure, showing how the chemical accidents happened. One hundred one chemical accidents were analyzed through the extended Human Factors Analysis and Classification System (HFACS) containing manifestations, and the associations between the high-frequency manifestations of adjacent levels were evaluated using Fisher's exact test. 21 groups of significant associations were identified, which clearly show how the manifestations at the upper level of the HFACS predict the adjacent lower level manifestations. A variety of potential failure paths for chemical accidents were presented, which need to be given more attention in developing future preventive measures. This work provides a new perspective and a theoretical basis for the prevention of chemical accidents.