In search engines, though feature-based query is provided, Content Based Image Retrieval (CBIR) still results in less sensitivity and specificity. It is because the conventional approach is based on feature extraction and inherent parameters in conventional feed forward networks. Performance of the system is strongly dependent on the extracted features. Hence it is necessary to develop a CBIR system that retrieves the similar images without explicit feature extraction. Convolutional Neural Networks are the recent neural network architectures which accept images as input and perform both feature extraction and classification. The proposed work aims at using the conventional architectures of VGGNET, RESNET, and DENSENET for flower classification in CBIR. Performance is measured in terms of accuracy of classification.