PURPOSE To find magnetic resonance (MR) findings predicting cement leakage in patients receiving percutaneous vertebroplasty (PVP) due to osteoporotic compression fractures. MATERIAL AND METHODS MR was done in 43 patients (age 52-89 years) before PVP (56 vertebrae), which was done via a bipedicular approach with fluoroscopic monitoring. Shortly after the procedure, a non-contrast-enhanced computed tomography (CT) scan was done at the vertebroplasty sites for evaluation of bone cement leakage. The following MR findings of fractured vertebral bodies were retrospectively reviewed for correlation with leakage risk: severity of bone-marrow edema, presence of a T2-weighted low-signal-intensity line, percentage residual vertebral body height, presence of either a vacuum or cystic change within the vertebral body, and presence of a cortical disruption. RESULTS After PVP, cement leakage was detected in 35 vertebrae (62.5%). Leakage was most frequently observed in the anterior external vertebral venous plexus (46%). Cement leakage rate increased when there was a cortical disruption (P = 0.037), especially at the endplates, while it decreased when there was a vacuum or a cystic change within the fractured vertebra (P = 0.019). Other MR findings were not related to the risk of cement leakage. CONCLUSION The risk of cement leakage in PVP increases when MR shows cortical disruption in a fractured vertebral body, especially at the endplates. It decreases when MR shows a vacuum or cystic change within the body.