The advent of PET instrumentation signaled the beginning of a new perspective in nuclear medicine diagnostic imaging. PET systems rely on several corrections that must be applied in order to establish accurate and reliable quantification. The inherent properties of PET detector architecture and the crystals themselves are sources of different types of systematic and random errors with subsequent count rate variability that should be accounted for. Normalization is the correction dealing with these errors. In this work, the reasons resulting in this variability are explored and the different normalization approaches are described. Special focus is paid to component-based normalization, with an attempt to describe the discrete factors and discuss the underlying mechanisms of their contribution to sensitivity variations of the lines of response.