Deep learning-based methods for expression recognition achieve improved recognition results but increase the number of parameters, and the computational requirement is high. The key challenge is to address the problem of too many parameters and complex structure of the network. In this research work, an enhanced residual attention network is constructed to extract the most important features to recognize. The convolutional network comprises residual blocks, maximum pooling layers, a spatial attention mechanism, and a SoftMax classifier. Spatial attention mechanism makes the convolutional model focus on more important regions. The proposed model improves the recognition rate and reduces the number of parameters, or computational complexity. The experimentation achieved an accuracy of 96.88% on the publicly available dataset of CK + including 1236 facial images. The proposed model proves that the result with attention improves 2.24%, which is better than without attention. The results demonstrate that the proposed model achieves competitive or better accuracy compared with the existing researches.