Golden Angle radial sampling, where each successive radial arm is rotated by 111.246°, is appealing for dynamic imaging because it has reduced sensitivity to motion and provides uniform k-space coverage for any arbitrary number of consecutive projections [1]. A balanced steady state free precession (bSSFP) sequence with Golden Angle profile ordering often results in image artifacts caused by gradient delays and varying eddy currents during the rapidly switching gradient scheme [2]. In recent work, the gradient system impulse response function (GIRF) has been characterized and used to predict true k-space trajectories during spiral and EPI imaging to reduce image distortion [3, 4]. Here, we propose to store a history of the time-varying gradient waveforms during a Golden Angle acquisition scheme and apply the GIRF calibration to predict true k-space coordinates during sampling in order to reduce image artifacts. Fully sampled Golden Angle radial bSSFP imaging was performed at 1.5T (Aera, Siemens Healthcare, Erlangen, Germany). The GIRF was calibrated for each gradient axis (GIRF bandwidth = 100 kHz and resolution = 26 Hz) [4] and convolved with the nominal x, y and z gradient waveforms in order predict the true gradient waveforms for arbitrary slice orientation. Images were reconstructed in MATLAB (R2013a, Mathworks, Natick, MA) with a nonuniform FFT. Three images were compared: 1) assuming nominal gradient waveforms for perfect radial projections, 2) using the GIRF-predicted k-space trajectories treating each radial profile independently (including the prephasing and readout gradients) and 3) using the GIRF-predicted k-space trajectories informed by the entire gradient waveform history up to and including the current profile. Phantom imaging and in vivo imaging of a swine heart were performed using the following parameters: TE/TR = 1.47/2.94 ms, 201 radial projections, FOV = 300mm, flip angle = 45°, matrix = 128 x 128, slice thickness = 5 mm, bandwidth = 1000 Hz/Px. Animal experiments were approved by the institutional animal care and use committee according to contemporary NIH guidelines. The improvement in image artifact was visible in both an axial phantom image (Figure 1) and an oblique short axis image of the swine heart (Figure 2). Images reconstructed using the nominal radial trajectories displayed significant image streaking and shading (Figure 1A, 2A). Images were improved by applying GIRF prediction to calculate k-space trajectories from each radial profile independently, but some residual image artifact remained (Figure 1B, 2B). Using the entire gradient waveform history to inform k-space trajectory prediction produced the best quality images, with negligible remaining artifact (Figure 1C, 2C). The images shown here were in transition to steady state, creating worst-case scenario artifacts. Figure 3 shows the trajectories through the center of k-space for the final 6 radial projections, where the nominal trajectories assume perfect central alignment of the radial profiles and the GIRF-predicted trajectories depict a mismatch at the k-space center caused by both gradient delays and eddy currents. Using the GIRF calibration for a retrospective trajectory prediction can simultaneously correct distortions from both eddy currents and gradient delays, and can be applied as a standard step during real-time image reconstruction [4], which could improve the clinical applicability of radial imaging. In the future, this method will be implemented in real-time using a buffer of the gradient history and we will apply this distortion correction during MRI-guided interventions where multiple slice orientations are interleaved, thus amplifying the eddy current artifacts. Further studies will assess the duration of waveform history required for sufficient artifact correction. Our preliminary data suggests that the ability to store a gradient waveform history allows a better prediction of the sampled k-space locations with the gradient system impulse response function.