In general, the historical data recording for natural disasters in small region is not enough to be used for estimating the probability distribution in risk estimation, because the size of the observation sample is small. In this paper, the information diffusion method is used to change single sample observations into fuzzy sets, and a quantitatively analyzing model for agricultural drought risk is proposed. The model is applied in a case study in Minqin county of Gansu province. The results show that the drought frequency in study region is high, but the degree of agricultural loss is low. Most of the losses are below 20%, and the losses with degree more than 30% have little possibility. Even so we cannot ignore the drought risk, because the possibility of agriculture losses at more than 10% degree still reaches 76%. The simulating result is proved to be accordant with the natural phenomena. This research gives a new way of risk analysis using information diffusing principle.