Abstract: In order to analyze reliability of loess high slopes, 266 natural ultimate state slopes are measured in Gansu area. According to the characteristics of topography and lithology, the studied region is divided into 4 subareas. The correlations of ultimate state slope height and width are given by means of double log-linear regression in different confidences for the 4 subareas respectively. Meanwhile, 1 024 groups of physico-mechanical indices obtained from reports are statistically analyzed. Based on the above, the ultimate state equation is built by means of Bishop's method. Reliability analysis for slopes under different confidences in 4 subareas is carried out with the Monte Carlo method. The effects of slope height, slope ratio and variability of parameters on the probability of failure are systematically analyzed. The results show that the slope ratio in Lintao-Yongjing subarea is smaller than 31.5? if the probability of failure(Pf) is kept within 10%. In the condition of the average coefficient of variation in Longdong subarea, Qin’an Lanzhou-Dingxi-Huining subarea and Tianshui-Qin’an-Tongwei subarea, the maximum of Pf of loess slopes is less than 10% with safety factor(Fs) of 1.3, as well as Pf less than 20% with Fs =1.2.

Key words: loess slope, Monte Carlo method, reliability, failure probability, ultimate state slopes