Diffusion tensor imaging (DTI) has seen increased usage in clinical and basic science research in the past decade. By assessing the water diffusion anisotropy within biological tissues, e.g. brain, researchers can infer different fiber structures important for neural pathways. A typical DTI data set contains at least one base image and six diffusion-weighted images along non-collinear encoding directions. The resultant images can then be combined to derive the three principal axes of the diffusion tensor and their respective cross terms, which can in turn be used to compute fractional anisotropy (FA) maps, apparent diffusion coefficient (ADC) maps, and to construct axonal fibers. The above operations all assume that DTI images along different diffusion-weighting directions for the same brain register to each other without spatial distortions. This assumption is generally false, as the large diffusion-weighting gradients would usually induce eddy currents to generate diffusion-weighting direction-dependent field gradients, leading to mis-registration within the DTI data set. Traditional methods for correcting magnetic field-induced distortions do not usually take into account these direction-dependent eddy currents unique for DTI, and they are usually time-consuming because multiple phase images need to be acquired. In this report, we describe our theory and implementation of an efficient and effective method to correct for the main field and eddy current-induced direction-dependent distortions for DTI images under a unified framework to facilitate the daily practice of DTI acquisitions.