Trademarks are valuable assets that need to be protected from infringement for the sake of producers and consumers. Therefore, Trademark Image Retrieval (TIR) is getting an increasing attention both academically and commercially. Recently, convolutional neural networks have stand out as a compulsory alternate. It offers perfect predictive performance and the possibility to replace classical workflows with an only network architecture. In addition, the transfer learning can save time and efforts in building deep convolutional neural networks. In this study, a transfer learning based TIR system is presented. It employs AlexNet, a pre-trained deep convolutional neural network. The proposed system is evaluated and validated using the two benchmark datasets: "FlickrLogos32" and "Logos-32 Plus" in terms of well-known performance metrics. The obtained results show that our proposed system has a promising performance compared to other recent systems.