In recent years, SN ratio has come into use in areas of design and development. It is desirable as the SN ratio is defined to find the solution with the smallest scatter. However, it can not be applied to every case. In design of parameters by Taguchi method, decision based on the SN ratio may cause some problems. Observation of experiment varies due to the sampling error even if it is sampled from the same population. And the SN ratio calculated from samples varies, too. When judged only by the SN ratio of the experiment result, it has the danger to end up unacceptable as the desired optimal condition. The failure to locate the optimal condition can be prevented when the amount of influence to the SN ratio by sampling error is estimated quantitatively. We analyze the characteristics of the behavior of SN ratio, and propose an improved procedure to determine the optimal parameters.