Denoising Diffusion Recommender Model (DDRM) is proposed to enhance the robustness of user and item embeddings in recommender systems against noisy implicit feedback. DDRM leverages the multi-step denoising process of diffusion models to improve embedding robustness. The model injects controlled Gaussian noises in the forward process and iteratively removes noises in the reverse denoising process. To guide the reverse denoising process, DDRM incorporates collaborative information as denoising guidance. During inference, DDRM uses the average embeddings of users' historically liked items as the starting point instead of pure noise, which increases the difficulty of the denoising process. Extensive experiments on three datasets with three representative backend recommender models demonstrate the effectiveness of DDRM. DDRM outperforms other baselines in terms of recommendation performance, particularly in noisy settings. The model's effectiveness is attributed to its robust representation learning through the diffusion process. DDRM is model-agnostic, allowing it to be deployed on any recommender model with user and item embeddings. The model's performance is evaluated using metrics such as Recall@K and NDCG@K. The results show that DDRM consistently outperforms other methods, including model-agnostic denoising methods and generative methods. The model's design variations, including different diffusion steps and noise schedules, are analyzed to validate its performance. The results indicate that DDRM is more efficient than traditional diffusion models in terms of inference time and computational cost. The model's effectiveness is further supported by ablation studies and hyper-parameter analysis, which show that the reconstruction loss and reweighted loss contribute significantly to DDRM's performance. Overall, DDRM demonstrates superior performance in denoising implicit feedback and enhancing recommendation accuracy.Denoising Diffusion Recommender Model (DDRM) is proposed to enhance the robustness of user and item embeddings in recommender systems against noisy implicit feedback. DDRM leverages the multi-step denoising process of diffusion models to improve embedding robustness. The model injects controlled Gaussian noises in the forward process and iteratively removes noises in the reverse denoising process. To guide the reverse denoising process, DDRM incorporates collaborative information as denoising guidance. During inference, DDRM uses the average embeddings of users' historically liked items as the starting point instead of pure noise, which increases the difficulty of the denoising process. Extensive experiments on three datasets with three representative backend recommender models demonstrate the effectiveness of DDRM. DDRM outperforms other baselines in terms of recommendation performance, particularly in noisy settings. The model's effectiveness is attributed to its robust representation learning through the diffusion process. DDRM is model-agnostic, allowing it to be deployed on any recommender model with user and item embeddings. The model's performance is evaluated using metrics such as Recall@K and NDCG@K. The results show that DDRM consistently outperforms other methods, including model-agnostic denoising methods and generative methods. The model's design variations, including different diffusion steps and noise schedules, are analyzed to validate its performance. The results indicate that DDRM is more efficient than traditional diffusion models in terms of inference time and computational cost. The model's effectiveness is further supported by ablation studies and hyper-parameter analysis, which show that the reconstruction loss and reweighted loss contribute significantly to DDRM's performance. Overall, DDRM demonstrates superior performance in denoising implicit feedback and enhancing recommendation accuracy.