Abstract: Landslide displacement prediction is one of important parts of landslide disaster prevention and mitigation. To improve the accuracy and precision of landslide displacement prediction is emphasis and difficulty. Outliers from monitoring samples are took into account in this research. By ignoring, reserving or correcting outliers to study on which is the best of three ways of landslide displacement prediction with outliers. The Zhujiadian Landslide in Three Gorges Reservoir Region is chosen as the case study on displacement prediction. Based on ARIMA(p, d, q) model, predictions are carried out using the accumulated displacement and the displacement rate time series, respectively. The research results show that: (1) The landslide prediction result with correcting outliers is between ignoring and reserving ones. (2)For ARIMA model, it is more suitable for using the displacement rate time series. (3)The prediction results of "step-like" displacement rates based on ARIMA(1, 0, 1) model for the displacement rate time series with correcting outliers are 79.0mm and 70.2mm in Jun. 2016 and 2017, and accumulated displacement is 1647.7mm until Aug. 2017.