The principal difficulty in implementing a code search engine is the difference in syntax between the natural-language query and computer-language target. Because of this, engines based on traditional IR techniques often have difficulty returning relevant code snippets. In this talk we discuss DeepCS, presented by Gu et al at ICSE last year, which uses a deep learning based model to map method definitions and their corresponding textual descriptions to nearby locations in the same feature space. In so doing, this system is also able to map natural-language queries to points in this feature space close to relevant code. This deep learning-based technique performs significantly better than Lucene-based systems, and even out-performs the state-of-the-art system CodeHow.