Cervical cancer remains one of the leading causes of mortality among women worldwide, emphasizing the urgent need for accurate and early diagnosis. This study proposes a deep learning–based framework for the automatic classification of cervical lesions from colposcopic images using Convolutional Neural Networks (CNNs). The dataset used was the Intel ODT for Cervix Images, preprocessed to enhance contrast and reduce noise variability. Three CNN architectures—VGG16, Xception, and InceptionV3—were trained and evaluated to determine the optimal configuration for lesion classification. Among them, InceptionV3 achieved the best individual performance, reaching an accuracy of 0.90, precision of 0.94, recall of 0.85, and a weighted F1-score of 0.82. To further enhance robustness, an ensemble learning approach combining the three models through a weighted majority voting scheme improved the overall F1-score to 0.86 and achieved a more balanced sensitivity across lesion types. The results demonstrate the potential of CNN-based ensembles as reliable diagnostic support tools for computer-aided cervical cancer screening, contributing to early detection and improved clinical decision-making.