Landslide is one of the natural disasters are common in Indonesia because the country is in the ring of fire. Landslides can be predicted by the empirical modeling of threshold rainfall that are used as early warning system. In order to model of threshold can be used as early warning system of landslide, it is necessary to validate to know the accuracy level of rainfall threshold in predicting the incidence of rainfall that triggered landslides and not a landslide. In this study, modeling was done with model empirical (intensity – duration/ID) approach using the 220 data of rainfall that triggered landslide with satellite-based TRMM in the territory of Indonesia. The intensity of the antecedent rainfall and the duration of antecedent rainfall used in modelling of rainfall threshold. Rainfall threshold is validated with ROC analysis. This method uses seven index statistic and ROC curve to determine the accuracy rate of the rainfall threshold. The results obtained empirical equation I = 9.65D-0.505 with value of rainfall intensity prediction generated increasingly low even close to zero in the interval time 1 – 30 days. The results of the analysis of the ROC on the rainfall threshold is indicate that the model has a pretty good accuracy rate and can be used as early warning system of landslide even though it still has a fairly large error rate.