DiffLinker is an E(3)-equivariant 3D conditional diffusion model for molecular linker design. It connects disconnected molecular fragments to generate chemically relevant drug molecules. Unlike previous methods that only connect pairs of fragments, DiffLinker can link an arbitrary number of fragments. It automatically determines the number of atoms in the linker and their attachment points. The model outperforms other methods in generating diverse and synthetically accessible molecules. It can also be conditioned on target protein pockets to generate linkers that are structurally compatible with the corresponding pockets. DiffLinker is equivariant to translations, rotations, reflections, and permutations, and it does not require information on the attachment atoms or a predefined linker size. It is also capable of generating linkers with varying sizes. DiffLinker has been tested on several benchmarks, including ZINC, CASF, and GEOM datasets, and has shown superior performance in terms of chemical relevance, synthetic accessibility, and drug-likeness. It has also been shown to generate more diverse linkers than other methods. DiffLinker has been applied to real-world drug design applications, including fragment-based design of Hsp90 and IMPDH inhibitors, and scaffold hopping for improving selectivity for JNKs. The model has been shown to generate linkers that are structurally compatible with protein pockets and to improve the binding affinity of generated compounds. DiffLinker is a versatile and state-of-the-art generative method applicable to various structure-based drug design tasks.DiffLinker is an E(3)-equivariant 3D conditional diffusion model for molecular linker design. It connects disconnected molecular fragments to generate chemically relevant drug molecules. Unlike previous methods that only connect pairs of fragments, DiffLinker can link an arbitrary number of fragments. It automatically determines the number of atoms in the linker and their attachment points. The model outperforms other methods in generating diverse and synthetically accessible molecules. It can also be conditioned on target protein pockets to generate linkers that are structurally compatible with the corresponding pockets. DiffLinker is equivariant to translations, rotations, reflections, and permutations, and it does not require information on the attachment atoms or a predefined linker size. It is also capable of generating linkers with varying sizes. DiffLinker has been tested on several benchmarks, including ZINC, CASF, and GEOM datasets, and has shown superior performance in terms of chemical relevance, synthetic accessibility, and drug-likeness. It has also been shown to generate more diverse linkers than other methods. DiffLinker has been applied to real-world drug design applications, including fragment-based design of Hsp90 and IMPDH inhibitors, and scaffold hopping for improving selectivity for JNKs. The model has been shown to generate linkers that are structurally compatible with protein pockets and to improve the binding affinity of generated compounds. DiffLinker is a versatile and state-of-the-art generative method applicable to various structure-based drug design tasks.