We present a component-based normalization method for a PET scanner with a rotating array of panel detectors. The normalization procedure is designed to robustly normalize sinograms acquired with different levels of scattered radiation and at different count rates, flattening the response and removing artifactual detector-block patterns, which if uncorrected could cause ring artifacts in images and intensity variations along the scanner axis. Coincidence events are binned during acquisition into a 3-D sinogram S(r,/spl phi/,z,seg) where r is radius, /spl phi/ is angle, z is the axial position, and seg is the segment. In its simple form, the normalization N, defined by the equation S(normalized)=N*S(unnormalized), is assumed to take the angle-independent form N=1/(g(r,seg)*f(r,z,seg)*c(z,seg)), where the f term contains only low radial frequencies. The g and f terms are derived from a scan of a scatter-free uniform source. The c term, derived from a scan of a cylindrical phantom that scatters radiation and fills the axial field of view, removes high-frequency axial features in the sinogram but does not correct for the low-frequency scatter features which are handled by the reconstruction software. As the count rate changes, one finds that sinograms normalized by this simple procedure are afflicted by a high-frequency axial pattern due to a count rate dependent sensitivity change at the edges of blocks. This artifact is addressed by a multiplicative correction of the form /spl gamma/=/spl gamma/(S,z), where S is the singles counting rate and /spl gamma/ is a periodical function with the periodicity of the array of detector blocks.