To assess the feasibility and accuracy of 3D ultrashort echo time (UTE) magnetic resonance (MR) thermometry of frozen tissue during cryoablation on a clinical 3T MR system.Ex vivo porcine muscle specimens (n?=?4) were imaged during two cycles of 10:3 minutes freeze-thaw on a 3T clinical MR scanner. Continuous MR monitoring was performed using a 3D radial ramp-sampled UTE sequence with a shortest TE of 70 ?s. Fiber optic sensors were used for temperature reference. Data of three experiments were used as reference sets. Signal intensity values were normalized to baseline before cooling and related to temperature. Data for subzero temperatures were fit to a monoexponential function. In the separate validation set, the obtained fit was used to generate 3D MR temperature maps of frozen tissue at each imaging timepoint. Statistical analysis was performed to assess accuracy of the MR temperature maps.With 3D UTE imaging, MR signal was measured from frozen tissue down to temperatures as low as -40Ã¯Â¿Â½C. Temperatures predicted from the MR temperature maps strongly correlated with sensor recorded values (r?=?0.977, P < 0.001). Bland-Altman analysis demonstrated a mean difference between MR-estimated temperatures and sensor readings of -1.2?Ã¯Â¿Â½?2.7Ã¯Â¿Â½C with upper and lower limits of agreement of +4.1 and -6.5Ã¯Â¿Â½C, respectively.3D MR thermometry of frozen tissue using UTE signal intensity was feasible during cryoablation on a clinical 3T MR system. Down to temperatures as low as -40Ã¯Â¿Â½C, accuracy of the MR temperature maps was within clinically acceptable limits.J. Magn. Reson. Imaging 2016.