With the development of oil and gas exploration and the updating of marine geophysical algorithms, the ocean bottom node (OBN) technology has emerged due to its advantages of converted waves, symmetric acquisition, wide azimuths, and long offset data. However, routine QC of OBN often requires manual selection of anomalous nodes from tens of thousands of nodes as the basis for subsequent correction processing. In this paper, an unsupervised intelligent QC method for OBN data based on an Auto-Encoder and Gaussian Mixture Model is proposed. The unsupervised OBN data intelligent QC task is considered as a clustering-based anomaly detection problem. Concretely, the distribution of OBN anomaly node data relative to standard node data in the feature domain is different. Therefore, the clustering centers of the standard node data, which are in absolute majority, can be estimated by clustering all the input OBN data. Described in detail, firstly, low-dimensional features and reconstruction errors are extracted from the average first-break amplitude of each OBN node using a deep auto-encoder and further transferred into the Gaussian mixture model. Second, expectation-maximizing joint optimization of the parameters for both the auto-encoder and the Gaussian model is performed in an end-to-end approach to balance the reconstruction error, implicit density estimation, and parameter regularization. Finally, the anomaly detection for OBN data is quantitatively controlled by calculating the maximum likelihood energy value. This method automates and intellectualizes OBN data QC by exploiting the advantages of deep learning in big data analytics. In addition, the deep learning method compresses the manual work period from weekly to minute level. The method based on unsupervised learning also avoids the problems of supervised learning manual labeling cost and the generalization ability of abnormal uncertainty in the fieldwork area.