MolCRAFT is a novel structure-based drug design (SBDD) model that operates in a continuous parameter space, addressing key challenges in generating molecules with accurate 3D conformations and binding affinity. The paper introduces MolCRAFT, the first SBDD model that works in a continuous parameter space, along with a novel noise-reduced sampling strategy. It addresses issues such as mode collapse and the hybrid continuous-discrete space in existing generative models for SBDD. Empirical results show that MolCRAFT consistently achieves superior performance in binding affinity with more stable 3D structures, demonstrating its ability to accurately model interatomic interactions. MolCRAFT achieves reference-level Vina Scores (-6.59 kcal/mol) with comparable molecular size, outperforming other strong baselines by a wide margin (-0.84 kcal/mol). The model is designed to handle the challenges of generating accurate molecular conformations and binding modes, including distorted geometry, inferior binding mode, and generation failure. MolCRAFT operates in a fully continuous parameter space, enabling lower input variance and smooth transformation towards the target distribution. The model's performance is evaluated on a dataset of protein-ligand pairs, showing high binding affinity, conformation stability, and drug-like properties. MolCRAFT also demonstrates high sample efficiency, generating more complete molecules at a significantly faster rate than existing methods. The model's success is attributed to its ability to capture the complex data manifold for molecules and its noise-reduced sampling strategy. The paper concludes that MolCRAFT is a significant advancement in SBDD, offering a more accurate and efficient approach to generating molecules with desired properties.MolCRAFT is a novel structure-based drug design (SBDD) model that operates in a continuous parameter space, addressing key challenges in generating molecules with accurate 3D conformations and binding affinity. The paper introduces MolCRAFT, the first SBDD model that works in a continuous parameter space, along with a novel noise-reduced sampling strategy. It addresses issues such as mode collapse and the hybrid continuous-discrete space in existing generative models for SBDD. Empirical results show that MolCRAFT consistently achieves superior performance in binding affinity with more stable 3D structures, demonstrating its ability to accurately model interatomic interactions. MolCRAFT achieves reference-level Vina Scores (-6.59 kcal/mol) with comparable molecular size, outperforming other strong baselines by a wide margin (-0.84 kcal/mol). The model is designed to handle the challenges of generating accurate molecular conformations and binding modes, including distorted geometry, inferior binding mode, and generation failure. MolCRAFT operates in a fully continuous parameter space, enabling lower input variance and smooth transformation towards the target distribution. The model's performance is evaluated on a dataset of protein-ligand pairs, showing high binding affinity, conformation stability, and drug-like properties. MolCRAFT also demonstrates high sample efficiency, generating more complete molecules at a significantly faster rate than existing methods. The model's success is attributed to its ability to capture the complex data manifold for molecules and its noise-reduced sampling strategy. The paper concludes that MolCRAFT is a significant advancement in SBDD, offering a more accurate and efficient approach to generating molecules with desired properties.