Purpose: To evaluate the value of subtraction images when using MRI to assess liver tumors treated with percutaneous ablation. Materials and Methods: Following percutaneous ablation of 35 liver tumors, two abdominal radiologists, blinded to outcomes, independently reviewed follow-up MRI examinations for tumoral enhancement suggestive of residual/recurrent tumor and rated their confidence level. After one year, the readers reviewed the same examinations with added subtraction images. Accuracy of the detection of residual/recurrent tumor and contrast-to-noise ratios (CNR; for tumoral enhancement-to-liver, tumoral enhancement-to-ablation zone, and ablation zone-to-liver) were calculated with and without subtraction images and compared using Wilcoxon signed rank test. Interobserver variability was computed using Kappa (kappa) statistics. Results: Residual/recurrent tumor was present in 8 (23.5%) of 34 tumors. Accuracy of detecting residual/recurrent tumor with subtraction images and interobserver agreement (kappa = 0.72, good) were better than accuracy of detecting residual/recurrent tumor and interobserver agreement (kappa = 0.57, moderate) of enhanced MR images without subtraction. Mean CNR of subtraction images was significantly higher than that of enhanced MR images for tumoral enhancement-to-liver (0.2 +/- 5 versus 11.6 +/- 14.4, P = 0.03), tumoral enhancement-to-ablation zone (10.1 +/- 12.5 versus 34.4 +/- 29.4, P = 0.02), and ablation zone-to-liver (11.8 +/- 13.3 versus 102.5 +/- 238.4, P = 0.03). Conclusion: When using MRI, subtraction images help both detect and exclude residual/recurrent tumor following percutaneous liver ablations. J. Magn. Reson. Imaging 2013;37:407413. (C) 2012 Wiley Periodicals, Inc.