Automated photo tagging has established itself as one of the most compelling applications of deep learning. While deep convolutional neural networks have repeatedly demonstrated top performance on standard datasets for classification, there are a number of often overlooked but important considerations when deploy- ing this technology in a real-world scenario. In this paper, we present our efforts in developing a large-scale photo-tagging system for Flickr photo search. We discuss topics including how to select the tags that matter most to our users, de- velop lightweight, high-performance models for tag prediction, and leverage the power of large amounts of noisy data for training. Our results demonstrate that, for real-world datasets, training exclusively with noisy data yields performance nearly on par with the standard paradigm of first pre-training on clean data and then fine-tuning. We advocate for the approach of harnessing user-generated data in large-scale systems.