This paper discusses how beliefs of a decision maker (DM) should be revised or updated for safety control of a large-complex plant. We evaluate the expected value of a loss in the plant which is caused by a safety-control action of the DM. There exist the following two kinds of mechanisms for belief revision : (i) combination rules for basic probability assignment functions, and (ii) updating rules based on the conditioning of belief functions. For each case, we give an optimal rule for minimizing the expected loss in the plant. The following three points are proven : (1) the “best” rule of belief revision for assuring plant safety does not always minimize the expected loss in the plant, (2) the optimality of a combination rule depends on the “type” of the safety-control policy (i.e., the safety-preservation type or the fault-warning type), and (3) the optimality of a conditioning rule, on the other hand, is irrespective of the type of the safety-control policy.