The standard method to monitor temperatures using Magnetic Resonance Imaging (MRI) within a human requires the acquisition of two MR images at two different time-points, which leads to unavoidable errors. In this thesis a physically and mathematically motivated method is presented to calculate temperature information using a single MR image. In homogeneous tissue with linear changing susceptibility the phase of the MR signal can be described as harmonic function. Using this harmonic behavior of the MR phase and the Laplace equation, a reference-less Proton Resonance Frequency (PRF) method can be presented in this thesis, which enables non-invasive temperature calculation on one phase image acquired during heating using MR within humans. For temperature calculation using this reference-less PRF method, a thin border has to be placed around the heating target area. Further on, temperature data in and around the thermal zone was analyzed using an exponential fit based on the solution of the Bioheat equation proposed by Pennes. Reference-less online temperature monitoring was tested on patients with and without breathing triggering or gating simultaneously with a thermo therapy of the liver. The results show that this method can be used to predict necrosis and as a safety tool during local thermal therapies. The reference-less method presented in this thesis is a promising post-processing tool for temperature monitoring of moving and deforming organs. In comparison to the standard reference subtraction method, the reference-less approach is more flexible for clinical use, eliminates the need for respiratory triggers, enables higher temporal resolution, and eliminates inter-scan motion artifacts as well as under- or overestimation of temperature induced by swelling/shrinking of heated tissue.