Higher-order statistics based approaches and signal sparseness based approaches have emerged in recent decades to resolve the underdetermined direction-of-arrival (DOA) estima-tion problem. These model-based methods face great chal-lenges in practical applications due to high computational com-plexity and dependence on ideal assumptions. This paper presents an effective DOA estimation approach based on a deep residual network (DRN) for the underdetermined case. We first extract an input feature from a new matrix calculated by stack-ing several covariance matrices corresponding to different time delays. We then provide the input feature to the trained DRN to construct the super resolution spectrum. The DRN learns the mapping relationship between the input feature and the spatial spectrum by training. The proposed approach is superior to exis-ting model-based estimation methods in terms of calculation ef-ficiency, independence of source sparseness and adaptive ca-pacity to non-ideal conditions (e.g., low signal to noise ratio, short bit sequence). Simulations demonstrate the validity and strong performance of the proposed algorithm on both overde-termined and underdetermined cases.