Abstract：The traditional method of slope stability analysis is safety factor，and its applications have long time，broad prospects and abundant experience. However，the weakness of safety factor is that it can not consider uncertainties of parameters and models，for which，especially uncertainty of parameters，has a great influence on slope stability. Reliability theory can effectively solve uncertain and correlative problems in slope system. However，reliability theory is complicated and short of abundant experience. Therefore，it is necessary to build dual index system of slope stability coupling safety factor and reliability. Most likely safety factor (mean safety factor) was calculated with average value of parameters，and reliability of slope was calculated with variability of safety factor. Supposing that safety factors obey normal distribution in different maximum possible safety factors and variation coefficients，a series of most likely safety factors and variation coefficients of safety factor were taken，and failure probabilities of slope Pf were calculated. In this way，a series of most likely safety factors and failure probabilities were obtained. Multiplied most likely safety factor by corresponding reliability，reduced most likely safety factor，reliability safety factor of slope was obtained. From limit state of reduced safety factor，critical failure probabilities and variation coefficients corressponding different maximum possible safety factors were calculated. According to different maximum possible safety factors and their failure probabilities，divisional chart of slope stability was drawn. Choosing representative safety factor and its reliability，dual index system of slope stability coupling safety factor and reliability was built，by which slope stability could be measured. As a research example，stability of Yancun landslide was assessed with dual index system of slope stability；and research result shows that dual index system of slope stability is feasible method. Estimation of variance with moment estimation was expounded in little sample size.
