Abstract. We study the diameter termination criterion for approximating linear continuous problems. It is assumed that only nonexact information about the problem is available. We evaluate the quality of the diameter termination criterion by comparing it with the theoretically best stopping condition. The comparison is made with respect to the cost of computing an e-approximation. Although the diameter termination criterion is independent of a particular problem, it turns out to be essentially equivalent to the theoretical condition. Optimal information and the best way of constructing an «-approximation are exhibited.