In recent years, it has been argued by some neuroanatomists that three-dimensional (3-D) counting approaches must be used in studies of neural systems, so that 'unbiased' counts of neurons can be obtained. By contrast, two-dimensional (2-D) cell-counting methods are said to be 'assumption-based' and to yield inaccurate results. Working from the premise that all scientific methodologies are assumption-based and suffer from inherent biases, the current review considers the relative strengths and weaknesses of 2-D versus 3-D counting approaches. This comparison is from the standpoint of predictive performance with respect to bias, variance and fidelity to the actual spatial arrangements of cells in the tissue under study. When these considerations are taken, together with the human resources that are required in using either methodology, 2-D methods offer more practical alternatives that might even provide more scientifically accurate estimates compared with their 3-D counterparts.