<p>Accurate classification of human blood cells plays a crucial role in disease diagnosis, health monitoring, forensic investigations, and medical research. However, conventional pathology-based blood cell analysis is time-consuming, requires expert interpretation, and is prone to subjectivity. These limitations highlight the need for automated, reliable, and efficient blood cell classification methods. Recent advances in deep learning have shown promise in addressing these challenges. In this study, a multi-scale capsule network, termed msCapsNet, is proposed for automated classification of peripheral blood cells. Unlike conventional Convolutional Neural Networks (CNNs), which often lose spatial and orientation information and require large training datasets, the proposed msCapsNet integrates a multi-scale feature extraction module that captures comprehensive features across different resolutions before passing them to the capsule network. This design enhances feature richness and robustness, particularly for complex pathological images. The effectiveness of the proposed method is evaluated on multiple publicly available datasets, including the Peripheral Blood Cell (PBC) dataset, White Blood Cell (WBC) classification dataset, Rabin-WBC dataset, Leukocyte Images for Segmentation and Classification (LISC) dataset, and the Acute Lymphoblastic Leukemia Image Database 2 (ALL-IDB2). The comparative experimental analysis of the suggested msCapsNet model shows better performance with 99.76% accuracy, 99.59% precision, 99.63% sensitivity, and 99.96% specificity using PBC dataset. Experimental results demonstrate that msCapsNet outperforms existing state-of-the-art approaches, highlighting its potential as a reliable tool for automated blood cell classification and clinical decision support.</p>