The integration of deep learning techniques into search heuristics presents a transformative approach to solving combinatorial optimization problems (COPs) and has revitalized interest in using deep neural networks (DNNs) in the field of optimization. This chapter explores application of DNNs in two state-of-the-art learning scenarios, namely controlling parameter values in search techniques and directly controlling decision making in heuristics. Despite challenges such as their black-box nature and resource-intensive training requirements, DNN-based methods are showing significant progress versus traditional Operations Research methods. We present methods and results from recent literature showing the current abilities of these techniques and provide a critical assessment and outlook for future research.