Aiming at feature evaluation and model optimization problem of remaining life prediction of rolling bearings,a method of feature evaluation and model optimization for bearing life is proposed. Based on the monotonicity and sensitivity evaluation of the bearing characteristics, this method quantitatively evaluates the bearing capacity tracking ability, and then selects the multidimensional feature sets that characterize the degradation of bearing performance. In order to reduce the influence of the redundant information between the multidimensional feature sets on the life prediction, multidimensional feature sets are clustered and filtered by using the Affinity propagation (AP) clustering method. In order to describe the characterization of the bearing degradation state uniformly after AP clustering, self-organizing feature map (SOM) is used to construct the bearing health index. Finally, the dual exponential model is optimized by adaptive chaos particle swarm optimization (ACPSO) to predict the remaining bearing life. The test shows that the method can accurately describe the bearing operating period and effectively predict the bearing remaining life.