Starting from a brief overview of the origin and nature of applied gradient field imperfections, this talk will focus on practical consequences of non-ideal gradient temporal and spatial properties. As dictated by Lentz’s law, rapid field changes by pulsed gradient coils necessarily induce currents in nearby conductive surfaces within the MRI system that counteract intended magnetic flux change by the gradients. Geometry, conductivity and proximity of these surfaces relative to the gradient coils determine the mix of spatial and temporal modes in resultant eddy currents. Certainly, active shielded gradient coil designs of the mid-1980s were a major advance to reduce eddy currents by greatly limiting gradient fields reaching conductive surfaces of the warm bore and cryo-shield. Comprehensive modeling of gradient, shim and magnet elements is essential for effective integration of subsystems into a properly functioning MRI system. Despite proper modeling and an integrated design, eddy currents will remain due to fundamental physics, but they are controlled to have acceptably small impact in most applications. Subject to maximum slew rate and amplitude constraints, pre-emphasis of the gradient amplifier drive waveform to temporally ‘overshoot’ thereby compensate for predicted eddy-current ‘under-shoot’ to achieve a near-ideal waveform. However, in more demanding applications such as single-shot echo-planar particularly in combination with diffusion-weighted imaging, eddy-currents can still introduce artifacts including: ghosting, blur, geometric warp/shear/scale/shift, and signal loss. Note, just as their eddy-current source, severity of these artifacts is spatially-variable. Uncompensated eddy currents that scale with diffusion pulse amplitude and vary with DW axis direction produce spatially misaligned DWI and yield artifact when combined as in trace-DWI, ADC, FA, etc. Image registration of individual b-value and direction DWIs to a common reference (e.g. b0) prior to subsequent mathematical combination mitigates these errors, and is often automatically performed via “in-line” routines.In the ideal world, gradients add incremental z-field that scales linearly with offset from isocenter. Of course, real-world physics dictate gradients are linear over only a limited range in the bore. For example, local gradient strength 5% to 10% higher than nominal at locations 15cm off center in R/L and A/P directions, and 15% to 25% weaker than nominal strength 15cm off center in S/I direction are common on modern clinical MRI systems. Since spatial encoding of local distance scales with actual gradient strength, local distances are proportionally in error leading to 3D geometric distortion. Fortunately, spatial distortion due to GNL varies smoothly in a predictable manner based on gradient coil design properties therefore GNL distortion is usually corrected using in-line 3D gradient (un)warp routines. For many routine applications high spatial fidelity may not be essential, but for image-guided biopsy/intervention and spatial conformal therapies, spatially accurate 3D images are crucial. Beyond automatic 3D spatial (un)warp correction of MR images using known GNL properties, diffusion-weighted images contain additional GNL error that is typically left uncorrected. Moreover, this error is unrelated to additional spatial distortion manifest on SS-EPI due to local B0-inhomogeneity. Instead, the error is in degree of diffusion weighting that varies spatially due to non uniform gradient strength. While DWI/DTI/DSI acquisitions are denoted by a scalar “b-value”, one should instead envision 3D spatial “b-maps” for each applied diffusion gradient direction. Since diffusion-weighting scales with the square of gradient amplitude, systematic GNL bias error in calculated ADC can often be 10% to 20% 15cm off center in AP/RL direction and 30% to 50% in SI direction. Mathematical formalism of GNL as a tensor, derived from gradient coil design coefficients allows one to appropriately derive local diffusion weighting and greatly mitigate GNL bias in ADC maps.