Abstract: The deformation and displacement of sudden loess landslides are small and the time of duration is short, which make early warning and forecast of landslides difficult. In order to explore a new way to predict the instability time of these landslides and reduce economic losses and casualties, four landslides in the Heifangtai area of Gansu Province in 2019 are taken as the research objects, and the deformation stage of landslide is determined with the improved tangent angle mode. A simplified cumulative calculation method based on the improved tangent angle is proposed. The SLO model is used to predict the instability time. The difference in the predicted results is analyzed from the speed change trend and disaster-causing mode. The results show that (1) the SLO model is of a certain feasibility in predicting the instability time of sudden loess landslides, and the predicted accuracy obtained from the tangent angle of 80° is the highest. (2) The simplified accumulative calculation performance using the tangent angle as the dividing index reduces the impact of data fluctuations on the predicted results and improves the predicted accuracy. (3) When the inverse velocity change trend is “concave”, the probability of early prediction is large.When the inverse velocity change trend is “convex”, the probability of early prediction is small. And when the inverse velocity change trend is linear, the prediction accuracy is relatively high. (4) The prediction effect of this model in loess fall landslides is better than that of loess flow landslides.