This paper presents a study of accuracy issues in thermal modeling of high power LED modules on system level. Both physical as well as numerical accuracy issues are addressed. Incorrect physical assumptions may result in seemingly correct, but erroneous results. It is therefore important to motivate the underlying key physical assumptions of a thermal model. In this paper thermal measurements are used to calibrate a computational fluid dynamics (CFD) model of a high power LED module model at a reference application condition, and to validate it at other application conditions.