Due to the interference of random noise and the degradation characteristics of rolling bearings， traditional model-data-fusion based remaining useful life （RUL） prediction method of rolling bearings might be affected. Thus， a novel RUL prediction method is proposed based on the fusion indictor and model-data-fusion to improve the accuracy of RUL prediction of rolling bearings. First， the principle component analysis and exponentially weighted moving average algorithm are used to fuse the multiple charactering features for constructing a monotonous fusion indictor. Then， a determining scheme for the first predicting time is built on 3σ criteria to trigger the RUL prediction process， which can avoid the invalidity of the prediction process. Lastly， the Rauch-Tung-Striebel smooth filter algorithm is embedded into the prediction model to reduce the random fluctuation and achieve the reliable RUL of rolling bearing. Simulated and experimental cases demonstrate the effectiveness of the proposed method and its superiority over the traditional model-data-fusion RUL prediction method.