Attributed graph are a crucial data structure in nature. Attributed graph clustering aims to partition graph nodes in an unsupervised manner. Existing methods are typically categorized into contrastive learning and non-contrastive learning approaches. However, both of these methods for deep clustering on attributed graphs have limitations: the former disturbs the graph information when constructing multiple views, which negatively impacts the original data embedding, while the latter fails to effectively mine useful information across multiple views. To address these challenges, we propose a deep graph clustering method, DCGC. Notably, DCGC constructs different views without the need for manual perturbation of graph data, thus avoiding semantic drift caused by Gaussian noise injection or edge masking. The dual-contrast mechanism of DCGC enables contrastive learning by aligning single-view positive and negative sample pairs with cross-view sample pairs. This process aggregates the graph’s original structure and latent attribute information, and adaptively adjusts the graph structure using clustering pseudo-label membership relationships to optimize the clustering process. Compared to traditional deep contrastive graph clustering methods, DCGC offers more efficient and harmonious data processing. Finally, experiments on five benchmark datasets demonstrate the superiority of the DCGC clustering algorithm.