Magnetic resonance imaging (MRI) is a versatile imaging modality that is widely used to assess the morphology, function and tissue characteristics of the human heart. However, evaluation of the fetal heart relies principally on ultrasound, as fetal cardiac MRI is limited by the challenges associated with imaging a small, rapidly beating heart that is subject to various regular and spontaneous movements. These challenges are overcome by acquiring rapid 2D MR images, each providing limited view of the fetal heart in utero. The 2D images are then combined in a single 3D/4D reconstruction using unsupervised motion estimation, outlier detection and super-resolution imaging to compensate for fetal-maternal movements and spatio-temporal limits of MRI. This computational framework was applied to data collected in over 100 human fetal subjects, showing good agreement with echocardiography and improved diagnostic utility over 2D imaging for screening of structural cardiac anomalies in utero.