Conversational search enables complex information retrieval by accurately understanding the user’s actual search intent through multiple interactions between the user and the system. Existing conversational query rewriting methods are difficult to optimize directly for downstream tasks. Traditional conversational dense retrieval methods directly utilize the entire session as input, which introduces redundant noise from irrelevant conversation turns. To address these limitations, we propose the Generative Query Augmentation with Dual-View Contrastive Learning for Conversational Dense Retrieval (GQADCR). Initially, we leverage the fact that the response for each conversation turn is available during training and use large language models to generate multiple standalone search queries. Subsequently, using the Best-of-N sampling strategy, we select the optimal rewritten query based on the actual retrieval performance. Finally, we propose a novel dual-view contrastive learning strategy that aligns the representation of the conversational session with those of self-contained rewritten queries and target retrieval passages, thereby facilitating the training of dense retrievers. Extensive experiments on two public datasets demonstrate the effectiveness, efficiency, and generalizability of our approach (Our source code is available at https://github.com/ywy5516/GQADCR ).