Mounting evidence indicates that scatter and attenuation are major confounds to objective diagnosis of brain disease by quantitative SPECT. There is considerable debate, however, as to the relative importance of scatter correction (SC) and attenuation correction (AC), and how they should be implemented. The efficacy of SC and AC for 99mTc brain SPECT was evaluated using a two-compartment fully tissue-equivalent anthropomorphic head phantom. Four correction schemes were implemented: uniform broad-beam AC, non-uniform broad-beam AC, uniform SC + AC, and non-uniform SC + AC. SC was based on non-stationary deconvolution scatter subtraction, modified to incorporate a priori knowledge of either the head contour (uniform SC) or transmission map (non-uniform SC). The quantitative accuracy of the correction schemes was evaluated in terms of contrast recovery, relative quantification (cortical:cerebellar activity), uniformity ((coefficient of variation of 230 macro-voxels) x 100%), and bias (relative to a calibration scan). Our results were: uniform broad-beam (mu = 0.12 cm(-1)) AC (the most popular correction): 71% contrast recovery, 112% relative quantification, 7.0% uniformity, +23% bias. Non-uniform broad-beam (soft tissue mu = 0.12 cm(-1)) AC: 73%, 114%, 6.0%, +21%, respectively. Uniform SC + AC: 90%, 99%, 4.9%, +12%, respectively. Non-uniform SC + AC: 93%, 101%, 4.0%, +10%, respectively. SC and AC achieved the best quantification; however, non-uniform corrections produce only small improvements over their uniform counterparts. SC + AC was found to be superior to AC; this advantage is distinct and consistent across all four quantification indices.