Evaluation of a laboratory-scale microwave imaging system for non-invasive temperature monitoring has previously been reported with good results in terms of both spatial and temperature resolution. However, a new formulation of the reconstruction algorithm in terms of the log-magnitude and phase of the electric fields has dramatically improved the ability of the system to track the temperature-dependent electrical conductivity distribution. This algorithmic enhancement was originally implemented as a way of improving overall imaging capability in cases of large, high contrast permittivity scatterers, but has also proved to be sensitive to subtle conductivity changes as required in thermal imaging. Additional refinements in the regularization procedure have strengthened the reliability and robustness of image convergence. Imaging experiments were performed for a single heated target consisting of a 5.1 cm diameter PVC tube located within 15 and 25 cm diameter monopole antenna arrays, respectively. The performance of both log-magnitude/phase and complex-valued reconstructions when subjected to four different regularization schemes has been compared based on this experimental data. The results demonstrate a significant accuracy improvement (to 0.2 degrees C as compared with 1.6 degrees C for the previously published approach) in tracking thermal changes in phantoms where electrical properties vary linearly with temperature over a range relevant to hyperthermia cancer therapy.