The Human Connectome Project (HCP) is a 5-year initiative to map human brain connections and their variability in healthy adults. Using multiple imaging modalities, the project studies 1200 healthy adults, along with extensive behavioral and genetic data. The focus of this review is on diffusion MRI (dMRI) and structural connectivity. Recent advances in dMRI acquisition and processing allow high-quality in-vivo MRI data to be obtained while scanning a large number of subjects. These advances result from two years of optimization efforts during the project's pilot phase. The data quality and methods described here are representative of the datasets and processing pipelines that will be made freely available to the community starting in 2013. The HCP uses a Siemens 3T Skyra system with modifications to improve diffusion imaging, including a Siemens SC72 gradient coil and stronger gradient power supply. The system allows for high gradient amplitudes, which improve signal-to-noise ratio (SNR) and reduce repetition time (TR). Accelerated imaging is achieved using simultaneous multislice echo planar imaging (EPI) with multiband (MB) excitation and multiple receivers. This reduces TR and scanning time, with minimal SNR loss. The HCP also addresses gradient nonlinearities, which can affect diffusion-sensitising gradients, by using individual b values and gradient orientations for each voxel. Q-space sampling is optimized for angular coverage and fiber orientation estimation. Multi-shell schemes provide greater sensitivity in detecting fiber crossings than single-shell schemes. The optimal b values and their combination depend on available SNR and spatial resolution. The HCP uses a 1.25mm isotropic spatial resolution for datasets, which provides higher specificity and allows reconstruction of certain tract features not observed with lower resolution. The HCP also addresses distortion correction, using a model-based approach that simultaneously considers and corrects for susceptibility and eddy-current induced distortions, as well as head motion. The correction is based on manipulating acquisitions so that field inhomogeneities manifest differently in different images. A Gaussian Process (GP) predictor is used to estimate the corrected data. Fibre orientation estimation is performed using parametric deconvolution methods, such as the ball & stick model, which is extended for multi-shell datasets. Bayesian inference is used to estimate fibre orientations and their uncertainty, with automatic relevance determination (ARD) priors to determine the number of fibre compartments in each voxel. Tractography is performed using probabilistic streamline tractography, which propagates local fibre orientation information. Structural connectivity matrices are obtained using various seeding strategies, with examples shown in the HCP datasets. The HCP also utilizes GPUs to speed up computations, with significant speed-up factors achieved for Bayesian estimation of fibre orientations. The final dMRI protocol includes a Stejskal-Tanner pulsed gradient scheme, multiband factor of 3, and controlled aliasing PE shift. The protocol includes 3 shells at b=1000The Human Connectome Project (HCP) is a 5-year initiative to map human brain connections and their variability in healthy adults. Using multiple imaging modalities, the project studies 1200 healthy adults, along with extensive behavioral and genetic data. The focus of this review is on diffusion MRI (dMRI) and structural connectivity. Recent advances in dMRI acquisition and processing allow high-quality in-vivo MRI data to be obtained while scanning a large number of subjects. These advances result from two years of optimization efforts during the project's pilot phase. The data quality and methods described here are representative of the datasets and processing pipelines that will be made freely available to the community starting in 2013. The HCP uses a Siemens 3T Skyra system with modifications to improve diffusion imaging, including a Siemens SC72 gradient coil and stronger gradient power supply. The system allows for high gradient amplitudes, which improve signal-to-noise ratio (SNR) and reduce repetition time (TR). Accelerated imaging is achieved using simultaneous multislice echo planar imaging (EPI) with multiband (MB) excitation and multiple receivers. This reduces TR and scanning time, with minimal SNR loss. The HCP also addresses gradient nonlinearities, which can affect diffusion-sensitising gradients, by using individual b values and gradient orientations for each voxel. Q-space sampling is optimized for angular coverage and fiber orientation estimation. Multi-shell schemes provide greater sensitivity in detecting fiber crossings than single-shell schemes. The optimal b values and their combination depend on available SNR and spatial resolution. The HCP uses a 1.25mm isotropic spatial resolution for datasets, which provides higher specificity and allows reconstruction of certain tract features not observed with lower resolution. The HCP also addresses distortion correction, using a model-based approach that simultaneously considers and corrects for susceptibility and eddy-current induced distortions, as well as head motion. The correction is based on manipulating acquisitions so that field inhomogeneities manifest differently in different images. A Gaussian Process (GP) predictor is used to estimate the corrected data. Fibre orientation estimation is performed using parametric deconvolution methods, such as the ball & stick model, which is extended for multi-shell datasets. Bayesian inference is used to estimate fibre orientations and their uncertainty, with automatic relevance determination (ARD) priors to determine the number of fibre compartments in each voxel. Tractography is performed using probabilistic streamline tractography, which propagates local fibre orientation information. Structural connectivity matrices are obtained using various seeding strategies, with examples shown in the HCP datasets. The HCP also utilizes GPUs to speed up computations, with significant speed-up factors achieved for Bayesian estimation of fibre orientations. The final dMRI protocol includes a Stejskal-Tanner pulsed gradient scheme, multiband factor of 3, and controlled aliasing PE shift. The protocol includes 3 shells at b=1000