Continuous thermometry during a hyperthermic procedure may help to correct for local differences in heat conduction and energy absorption, and thus allow optimization of the thermal therapy. Noninvasive, three‐dimensional mapping of temperature changes is feasible with magnetic resonance (MR) and may be based on the relaxation time T₁, the diffusion coefficient (D), or proton resonance frequency (PRF) of tissue water. The use of temperature‐sensitive contrast agents and proton spectroscopic imaging can provide absolute temperature measurements. The principles and performance of these methods are reviewed in this paper. The excellent linearity and near‐independence with respect to tissue type, together with good temperature sensitivity, make PRF‐based temperature MRI the preferred choice for many applications at mid to high field strength (≥ 1 T). The PRF methods employ radiofrequency spoiled gradient‐echo imaging methods. A standard deviation of less than 1°C, for a temporal resolution below 1 second and a spatial resolution of about 2 mm, is feasible for a single slice for immobile tissues. Corrections should be made for temperature‐induced susceptibility effects in the PRF method. If spin‐echo methods are preferred, for example when field homogeneity is poor due to small ferromagnetic parts in the needle, the D‐ and T₁‐based methods may give better results. The sensitivity of the D method is higher that that of the T₁ methods provided that motion artifacts are avoided and the trace of D is evaluated. Fat suppression is necessary for most tissues when T₁, D, or PRF methods are employed. The latter three methods require excellent registration to correct for displacements between scans. J. Magn. Reson. Imaging 2000;12:525–533. © 2000 Wiley‐Liss, Inc.