MRI-derived thermometry based on the temperature-dependence of the proton resonant frequency (PRF) is extremely sensitive to changes in tissue unrelated to temperature changes, including tissue swelling. This study investigated the maximum amount of time that this phase-subtraction-based method can be used to accurately monitor temperature changes in vivo. Long-duration focused ultrasound sonications were delivered in rabbit thigh muscle with a phased-array transducer, and the time that tissue swelling began was monitored. Tissue swelling began to occur at about one minute. The temperature correlated well with an implanted thermocouple up to this time. After this time, severe artifacts in the phase-difference maps were observed. The thermal dose model predicted the extent of tissue damage well for subsequent one minute sonications. These results will have implications for MRI guidance of thermal therapies with long exposure times.