To measure a response, such as wave induced bending stress, for about 20-30 minutes is usual in full scale measurement, while in the case of measurement by scratch extreme stress gauge duration of it is 4 hours. In the latter case, they have never studied theoretically the type of distribution of extremal value of stress measured by the gauge. Authors study the problem from theoretical base. At first, authors pay attention to the group of peak value of stress measured for such longer time. The measured stress probability distribution consists of several group of distribution each of which is approximated by a Rayleigh distribution. By numerical analysis, it is found that the extremal distribution for such group is approximated by the Rayleigh distribution that have the largest parameter among the distributions. It means that an extreme distribution of the parameter has a significant effect on the long-term distribution of extreme values of a response. Usually the value of the parameter of the Rayleigh distribution, R , have a significant effect on the adjacent value of R . This means that a sample distribution of R is regarded as a distribution of samples drawn from a group, member of which subjects to a Markov chain. Authors make clear that the extreme distribution of R is approximated by a Weibull distribution by numerical simulation. It follows that a long-term distribution of extreme value of response is also expressed by a Weibull distribution approximately. By authors study, it is concluded that a result of measurement for such longer times is to be regarded as a sample from a population having a Weibull distribution.