Musculoskeletal Diseases (MSDs) are the major global health concern and the second leading cause of disability worldwide. Detecting MSDs via radiography is time-consuming and difficult due to the paucity of radiologists. The traditional Convolutional Neural Networks (CNNs) have been developed to help recognize MSDs, but they are computationally expensive, requiring millions of parameters and risking data loss due to pooling processes. To overcome these issues, Aggregated-FlexiNet for Interpretable Musculoskeletal Radiograph Classification is proposed. Initially, the input images are collected from two different datasets such as MURA dataset and FracAtlas dataset. Then, Contrast-Limited Adaptive Histogram Equalization (CLAHE) technique is employed to resizing the images. Afterwards, Fractional Max-Pooling (FMP) is used to reduce the information loss with overlapping or disjoint regions. Then, Aggregated-FlexiNet is used to predict the probability for a given radiograph of a specific study as shoulder, elbow, wrist, hand, finger, humerus and forearm. The experimental results demonstrate that the Aggregated-FlexiNet achieves superior performance compared to the existing models. The proposed Aggregated-FlexiNet achieves high Cohen’s Kappa value of 0.71 %, high sensitivity of 94.51 % and high specificity of 93.2 % when compared with existing models.