Cardiac magnetic resonance imaging (CMR) is affected by both cardiac and respiratory motion which are commonly compensated with ECG-gated imaging during a breath hold. However, free-breathing methods are preferred or even required in patients with limited breath hold capability. Free-breathing approaches utilize either rapid single-shot scans to reduce respiratory motion artifacts at the expense of spatial resolution or segmented respiratory-gated scans at the expense of scan time efficiency. Golden Angle radial [1] sampling has favorable properties for free-breathing CMR, including motion robustness and flexible reconstruction options. These properties also make Golden Angle radial sampling a good sampling strategy for motion compensated reconstructions [2,3]. In this work, we propose the combination of a self-gated 2D radial Golden Angle data acquisition scheme with a recently introduced motion compensated reconstruction strategy [4] to obtain high-resolution motion compensated CMR data in every desired cardiac phase.A 2D golden angle radial spoiled gradient echo sequence was employed to collect short-axis cardiac data from four volunteers using the following parameters: 3.0T scanner (MR750w, GE Healthcare, Waukesha/WI, USA), α=15°,BW=±125kHz,256 readout points,TR=4.3ms,TE=1.5ms,slice thickness=8mm, FOV=360mm circular, resolution 1.4x1.4mm², 2584 projections resulting in 11s total scan time. Self-gating signals for cardiac and respiratory motion were obtained from the k-space DC signal [5] and compared by regression to simultaneously recorded peripheral gating (PG) and respiratory belt signals. Using the derived self-gating signals, images in various cardiac and respiratory phases were reconstructed using a binning scheme as described previously, e.g., in [5] (Figure 1, step 1). For each cardiac phase, four images were reconstructed as follows: 1) Combining all free-breathing data per cardiac phase without motion management (NoMCR); 2) Retrospective self-gating selecting data (approx. 20-25%) closest to end-expiration of the desired cardiac phase (RG); 3) Motion Compensated Reconstruction (MCR); and 4) a filtered backprojection (FBP) of the RG data, which was viewed as ground truth to assess whether MCR accurately reconstructs the desired cardiac phase. NoMCR and RG recovered missing data using a non-Cartesian iterative SENSE [6]. The MCR was accomplished in four steps using the cardiac and respiratory self-gating signals: 1) clustering of all data of the desired cardiac phase into eight respiratory bins; 2) filtered backprojection of these eight respiratory states; 3) extraction of the calibration displacement fields between the bins by applying a non-rigid registration; 4) motion compensated SENSE-like reconstruction using the previously extracted calibration motion fields [4,7,8] (Figure 1). A very high correlation (R² = 95.9%) between the self-gating signals, the recorded PG curve, and respiratory belt was observed (Figure 2). Data from 11s scan time were found sufficient to result in high-quality motion compensated cine images. Figures 3 and 4 show representative images and temporal profiles of the four different reconstruction schemes. Combining all free-breathing data into one dataset (NoMCR) leads to high SNR images with significant respiratory motion induced blurring. RG improves apparent sharpness (e.g. diastole), but can suffer artifacts from severe undersampling (e.g. systole). MCR takes advantage of all projections in all respiratory states by including the calibration displacement fields into the reconstruction. Thus it recovers a sharp image with high SNR where small anatomical features are preserved. Furthermore, the cardiac phases in the MCR are in good agreement with the FBP ground truth images (Figure 4c), demonstrating the ability of the MCR to accurately recover the desired cardiac phases.Cardiac cine images from 11s fully self-gated, free-breathing Golden Angle radial datasets were successfully reconstructed using MCR. This scheme proved to be superior to non-Cartesian iterative SENSE reconstructions without motion correction (NoMCR, RG). The described technique offers several potential advantages: improved patient comfort and clinical workflow since neither breath holds nor ECG leads are necessary; high image quality with both high SNR and high spatial resolution; time-efficient acquisition since no data are discarded– instead, every acquired sample is used in the motion compensated image. The presented technique could also be extended to simplify and improve contrast-enhanced cardiac exams such as perfusion or late enhancement.