Functional magnetic resonance imaging (fMRI) is most commonly based on echo-planar imaging (EPI). With higher field strengths, gradient performance, and computational power, real-time fMRI has become feasible; that is, brain activation can be monitored during the ongoing scan. However, EPI suffers from geometric distortions due to inhomogeneities of the magnetic field, especially close to air-tissue interfaces. Thus, functional activations might be mislocalized and assigned to the wrong anatomical structures. Several techniques have been reported which reduce geometric distortions, for example, mapping of the static magnetic field B(0) or the point spread function for all voxels. Yet these techniques require additional reference scans and in some cases extensive computational time. Moreover, only static field inhomogeneities can be corrected, because the correction is based on a static reference scan. We present an approach which allows for simultaneous acquisition and distortion correction of a functional image without a reference scan. The technique is based on a modified multi-echo EPI data acquisition scheme using a phase-encoding (PE) gradient with alternating polarity. The images exhibit opposite distortions due to the inverted PE gradient. After adjusting the contrast of the images acquired at different echo times, this information is used for the distortion correction. We present the theory, implementation, and applications of this single-shot distortion correction. Significant reduction in geometric distortion is shown both for phantom images and human fMRI data. Moreover, sensitivity to the blood oxygen level-dependent (BOLD) effect is increased by weighted summation of the undistorted images.