Deep learning models achieve high accuracy in image recognition but often function as “black boxes”, making their decision-making processes difficult to interpret. Explainable AI (XAI) techniques aim to enhance transparency by providing insights into how deep neural networks reach their conclusions. This study presents a comparative evaluation of prominent XAI methods used in convolutional neural networks (CNNs), specifically Gradient-weighted Class Activation Mapping (Grad-CAM) and Saliency Maps. The techniques were applied to image classification tasks using benchmark datasets (ImageNet and CIFAR-10) and evaluated based on clarity, completeness, and trustworthiness. Our experimental results, conducted using VGG16 and ResNet50 architectures, demonstrate that Grad-CAM produces interpretable heatmaps that highlight relevant image regions, whereas Saliency Maps offer pixel-level feature importance with higher granularity but increased noise. The findings provide guidance for selecting suitable XAI methods depending on interpretability requirements, and we propose future research directions, including hybrid XAI approaches for improved transparency.