MRI phantom studies often fail to mimic the temperature of the human body, which can negatively impact accuracy. An artifact induced by increasing temperature in liquid phantoms was observed, presenting a significant challenge to temperature‐controlled experiments. In this study we characterize and provide a solution to eliminate this temperature‐induced MRI artifact. Low concentration (0.5–2.5 mM) agar phantoms were prepared. Utilizing a temperature‐controlled phantom holder, T1‐ and T2‐weighted structural images were acquired at 7 T along with quantitative B0, B1, T1, T2 and ADC maps at both 25 and 37°C. Additionally, computer simulations were conducted to demonstrate the fluid flow and thermal flux patterns in water to provide an insight into the origins of the artifact. Evidence from computer simulation and quantitative MRI strongly suggest the artifact was caused by heat transfer in the form of natural convection leading to structured patterns of signal loss in MR images. The artifact was present up to agar concentrations of 1.5 mM (T1 = 3068 ± 16 ms, T2 = 1052 ± 20 ms, ADC = 2.29 ± 0.36 × 10−3 mm2/s at 25°C; T1 = 3928 ± 44 ms, T2 = 1122 ± 24 ms, ADC = 2.64 ± 0.49 × 10−3 mm2/s at 37°C), above which point increased sample viscosity no longer allows for convection currents, thereby eliminating the artifact. The methodology described in this work simplifies quantitative MR acquisition of liquid phantoms at physiological temperature by suppressing convection currents with relatively small changes to intrinsic MR parameters (T1 increased by 1.4% and T2 decreased by 17% for 1.5 mM agar at 25°C).