We have been developing a risk recognition training tool for field overseers with VR technologies. At the first step, risk recognition knowledge required for the field overseers were elicited from a work procedures manual and from field overseers. Since the knowledge for risk recognition is regarded as a kind of tacit dimension of knowledge, it is difficult for the field overseers to verbalize them. A tool for promoting verbalization of the latent field overseers’ knowledge with the VR technologies, and an algorithm for structuring knowledge as propositional network were developed in the study.We elicited the risk recognition knowledge from 10 filed overseers in a construction company with the tool and structured the risk recognition knowledge with the algorithm based on the field overseers’ utterances. The risk recognition knowledge described in the work procedures manual was also structured with the algorithm, for compensating the knowledge required for the training tool. As the results, the differences between risk recognition knowledge described in the work procedure manual and that used in the real field are clarified. The knowledge described in the manual is mainly focused on direct causal relationship leading the immediate risk, which directly inflicts harm on workers. On the other hand, the knowledge elicited from the field overseers’ utterances is included indirect knowledge used to increase probability of the immediate risk.