Abstract: Predicting the remaining useful life (RUL) of rolling bearings before their degradation and failure is of great significance to ensure the safe operation of equipment and reduce economic losses. Therefore, aiming at the RUL prediction of rolling bearings, a new prediction flow was constructed based on multi-dimensional temporal mixing features. Firstly, a neural network based on one-dimensional convolutional neural network (1DCNN) and temporal convolutional network (TCN) was used to automatically extract the relevant features of the vibration signals. Secondly, multi-layer perceptrons were used alternately in the time and feature domain to construct a multi-dimensional temporal feature mixing model, and the features of the historical and the current moment were used as model inputs for the prediction of RUL. The experimental results show that the mean value of root mean square error and mean absolute error for the RUL prediction curve is reduced to 0.263 and 0.227, respectively, and the mean value of absolute error for failure point prediction is increased to 10.67%. Compared with the deep convolutional neural network and long short-term memory network, the proposed method has obvious advantages in both the fitting degree of the RUL prediction curve and the prediction of rolling bearings failure point. It can be seen that the RUL prediction flow of rolling bearings constructed can accurately predict their RUL and has certain practicability.