To implement noninvasive thermometry, we installed a hybrid system consisting of a radiofrequency multiantenna applicator (SIGMA-Eye) for deep hyperthermia (BSD-2000/3D) integrated into the gantry of a 1.5 Tesla magnetic resonance (MR) tomograph Symphony. This system can record MR data during radiofrequency heating and is suitable for application and evaluation of methods for MR thermography. In 15 patients with preirradiated pelvic rectal recurrences, we acquired phase data sets (25 slices) every 10 to 15 minutes over the treatment time (60-90 minutes) using gradient echo sequences (echo time = 20 ms), transformed the phase differences to MR temperatures, and fused the color-coded MR-temperature distributions with anatomic T1-weighted MR data sets. We could generate one complete series of MR data sets per patient with satisfactory quality for further analysis. In fat, muscle, water bolus, prostate, bladder, and tumor, we delineated regions of interest (ROI), used the fat ROI for drift correction by transforming these regions to a phase shift zero, and evaluated the MR-temperature frequency distributions. Mean MR temperatures (T MR ), maximum T MR , full width half maximum (FWHM), and other descriptors of tumors and normal tissues were noninvasively derived and their dependencies outlined. In 8 of 15 patients, direct temperature measurements in reference points were available. We correlated the tumor MR temperatures with direct measurements, clinical response, and tumor features (volume and location), and found reasonable trends and correlations. Therefore, the mean T MR of the tumor might be useful as a variable to evaluate the quality and effectivity of heat treatments, and consequently as optimization variable. Feasibility of noninvasive MR thermography for regional hyperthermia has been shown and should be further investigated. (Cancer Res 2005; 65(13): 5872-80)