White matter neuronal density has been correlated with clinical outcome after temporal lobectomy for refractory epilepsy. Both morphometric 2D (two-dimensional) and stereological 3D (three-dimensional) analyses of neuronal density have been performed. 3D analyses are thought to be more accurate than 2D counts, but more time-consuming. We compared 3D and automated 2D measurements in the same specimens. Adjacent 20-microm (for 3D analyses) and 5-microm (for 2D analyses) sections from 10 temporal lobectomies were stained for NeuN immunohistochemistry. Analysis of 100% of a region of interest (ROI) in deep white matter was performed using an image analysis system (Histometrix, Kinetic Imaging, UK). 3D analyses were undertaken using x 63 magnification (6 h/case). Automated 2D analyses were undertaken using automatic neuronal identification at x 10 magnification with three to four repeats (1.5 h/case). The range of neuronal densities for 3D measurements was 2120-4910 neurones/mm(3), and for automated 2D measurements 17.4-47.1 neurones/mm2. There was a linear correlation between the two methods with an r2 of 0.58. [corrected] Count-recount variability was 1.4-9.9% for the 3D and 5.1-36.6% for the automated 2D measurements. We found a wide range of white matter neuronal densities using either analysis. The low agreement between methods, and the high count-recount variability for the automated 2D analyses, indicate that despite being more time-consuming, rigorous 3D stereological analyses have to be performed to obtain reliable results. These findings have implications for studies requiring neuronal counts in normal and disease states.