Facial expression is a key feature of non-verbal communication. Facial expression recognition is a field within computer vision focused on enabling machines to interpret human emotions through facial expression. There are various approaches for achieving this, primarily involving either machine learning methods, such as Support Vector Machines (SVM) aided by feature extraction, or deep learning methods, such as Convolutional Neural Networks (CNN). An alternative deep learning approach, known as the Morphological Neural Network (MNN) has been developed, which leverages morphological operations for feature extraction instead of convolution. Previous studies have implemented opening morphological operations for MNN architectures, demonstrating that MNN can improve complexity efficiency while maintaining accuracy levels comparable to CNN. This study proposes a novel MNN architecture that utilizes a set of morphological operations within its morphological layer. The proposed MNN model is trained and tested on a comprehensive dataset that combines commonly used facial expression datasets, such as CK+, KDEF, TFEID, and JAFFE, along with an Indonesian facial expression dataset for additional diversity. This study compares the performance of the proposed MNN with previous MNN models and SVM with four different kernel functions Experimental results indicate that the proposed MNN outperforms both the existing MNN architecture and the SVM models.