It is generally well accepted that transmission (TX)-based non-uniform attenuation correction can supply more accurate absolute quantification; however, whether it provides additional benefits in routine clinical diagnosis based on qualitative interpretation of 3D brain positron emission tomography (PET) images is still the subject of debate. The aim of this study was to compare the effect of the two major classes of method for determining the attenuation map, i.e. uniform versus non-uniform, using clinical studies based on qualitative assessment as well as absolute and relative quantitative volume of interest-based analysis. We investigated the effect of six different methods for determining the patient-specific attenuation map. The first method, referred to as the uniform fit-ellipse method (UFEM), approximates the outline of the head by an ellipse assuming a constant linear attenuation factor (mu=0.096 cm(-1)) for soft tissue. The second, referred to as the automated contour detection method (ACDM), estimates the outline of the head from the emission sinogram. Attenuation of the skull is accounted for by assuming a constant uniform skull thickness (0.45 cm) within the estimated shape and the correct micro value (0.151 cm(-1)) is used. The usual measured transmission method using caesium-137 single-photon sources was used without (MTM) and with segmentation of the TX data (STM). These techniques were finally compared with the segmented magnetic resonance imaging method (SMM) and an implementation of the inferring attenuation distributions method (IADM) based on the digital Zubal head atlas. Several image quality parameters were compared, including absolute and relative quantification indexes, and the correlation between them was checked. The qualitative evaluation showed no significant differences between the different attenuation correction techniques as assessed by expert physicians, with the exception of ACDM, which generated artefacts in the upper edges of the head. The mean squared error between the different attenuation maps was also larger when using this latter method owing to the fact that the current implementation of the method significantly overestimated the head contours on the external slices. Correlation between the mean regional cerebral glucose metabolism (rCGM) values obtained with the various attenuation correction methods and those obtained with the gold standard (MTM) was good, except in the case of ACDM (R(2)=0.54). The STM and SMM methods showed the best correlation (R(2)=0.90) and the regression lines agreed well with the line of identity. Relative differences in mean rCGM values were in general less than 8%. Nevertheless, ANOVA results showed statistically significant differences between the different methods for some regions of the brain. It is concluded that the attenuation map influences both absolute and relative quantitation in cerebral 3D PET. Transmission-less attenuation correction results in a reduced radiation dose and makes a dramatic difference in acquisition time, allowing increased patient throughput.