Neural networks are increasingly able to outperform traditional machine learning and filtering approaches in classification tasks. However, with the rise in their popularity, many unknowns still exist when it comes to the internal learning processes of the networks in terms of how they make the right decisions for prediction. As a result, in this work, different attention modules integrated into a convolutional neural network coupled with an attention-guided strategy were examined for facial emotion recognition performance. A custom attention block, AGFER, was developed and evaluated against two other well-known modules of squeeze–excitation and convolution block attention modules and compared with the base model architecture. All models were trained and validated using a subset from the OULU-CASIA database. Afterward, cross-database testing was performed using the FACES dataset to assess the generalization capability of the trained models. The results showed that the proposed attention module with the guidance strategy showed better performance than the base architecture while maintaining similar results versus other popular attention modules. The developed AGFER attention-integrated model focused on relevant features for facial emotion recognition, highlighting the efficacy of guiding the model during the integral training process.