Summary Seismic data interpolation is a longstanding issue. Most of current interpolation methods are just suitable for random missing cases. However, seismic survey design using a random distribution of shots and receivers is always operationally challenging and impractical. For regular missing cases, anti-aliasing strategy can be included. In this abstract, we propose to use deep learning (DL) based residual networks (ResNets) for seismic data antialiasing interpolation which can extract high-level characteristics of training data by self-learning and avoid linear events, sparsity and low rank assumptions of traditional interpolation methods. Based on the convolutional neural networks (CNN), the 8 layer ResNets with better back propagation property for deep layers is designed for antialiasing interpolation. A set of simulated synthetic data is used to train the designed ResNets. The performance is assessed with several synthetic and field data, which demonstrates that the trained ResNets can help to reconstruct regularly missing traces with high precision. The interpolated results and the corresponding residuals demonstrate the validity of the trained ResNets.