Three-phase voltage unbalance in distribution network will have a greater impact on the loss of asynchronous motors. Using equivalent circuit formula to calculate motor loss under the influence of three-phase voltage unbalance has the problems of unstable accuracy, too many parameters and too complex mathematical model. In response to the above problem, an asynchronous motor loss assessment method based on the attention mechanism and convolutional neural network (CNN), namely Attention-CNN is proposed. This method takes the measured motor data as input, and introduces the attention mechanism to assign different weights to motor features. The CNN framework composed of the convolutional layer and the full connection layer is used to learn the measured data of the asynchronous motor, thus completing motor loss assessment. Taking the asynchronous motor loss data obtained from field experimentation as a practical example, the average error between the assessed loss of this method and the measured loss is only 0.717% and 0.549%. Comparing with the equivalent circuit and other typical machine learning algorithms, the proposed method has better performance in loss assessment showed in experimental results.