The remaining useful lifetime (RUL) of assets plays a critical role in machine prognostics and health management (PHM). Accurate RUL predictions can reduce losses caused by equipment faults. Most existing data-driven PHM methods rely on long short-term memory (LSTM) networks to model the relationship of time series data and RUL. However, because of the sequential nature of LSTM, it is not conducive to parallel computing. Herein, we propose the Deep & Attention Network, which uses a combination of convolutional neural networks and Attention methodologies instead of LSTM. In the proposed Deep & Attention Network, the Attention component models the temporal property, while the Deep component learns the effect of noise data. Experiments on NASA's Commercial Modular Aero- Propulsion System Simulation datasets demonstrate that the proposed network achieves a level of performance similar to that of other state-of-the-art RUL prediction models. Moreover, compared with LSTM-based methods, our Self-Attention-based method is conducive to parallel computing.