Purpose : To differentiate malignant vertebral compression fractures from benign fractures, as seen onspin-echo T1-weighted, fast spin-echo T2-weighted, and fat-suppressed gadolinium-enhanced T1-weighted MR images.Materials and Methods : Thirty two benign (18 acute and 14 chronic) and 28 malignant vertebral collapses werestudied in 54 patients aged between 15 and 78 (mean, 51) years. Malignant compression fractures involved onlymetastasis. We obtained sagittal and axial fast spin-echo T2-weighted images, and unenhanced and fat-suppressedgadolinium-enhanced T1-weighted images, and analyzed MR signal intensity, enhancement patterns, and morphologicchanges including convex posterior cortex vs retropulsion of a bone fragment, focal vs diffuse paraspinal mass,and epidural mass. Results : All cases of acute benign and malignant compression fractures showed low signalintensity within the vertebral body on T1-weighted images, and substantial contrast enhancement on fat-suppressedgadolinium-enhanced T1-weighted images. Acute benign and malignant compression fractures were distinguished on thebasis of three signal intensity characteristics: hypointense band (acute benign cases 77% ; malignant cases 0%),diffuse low signal intensity(17% vs 86%) and involvement of pedicle (0% vs 75%). Fast spin-echo T2-weighted imagesplayed little role in distinguishing between the two. Three morphologic changes were suggestive of malignancy:convex posterior cortex (malignant cases 75% ; benign 0%), epidural mass (79% vs 5%), and focal paraspinal mass(57% vs 0%). Retropulsion of a bone fragment (benign cases 63% ; malignant cases 14%) was preferable forbenignancy. in addition, thin diffuse paraspinal mass (benign cases 15% ; malignant cases 14%) was seen.Conclusion: The morphologic and signal intensity characteristics seen on T1-weighted images were useful fordistingushing benign and malignant vertebral compression fractures. Fast spin-echo T2-weighted and fat-suppressedgadolinium-enhanced T1-weighted images played little role in distinguishing between the two.