ERP researchers use differences in scalp distributions to infer differences in spatial configurations of neuroelectric generators. Since McCarthy and Wood (1985) demonstrated that a spatially fixed current source varying only in strength can yield a significant Condition × Electrode interaction in ANOVA, the recommended approach has been to normalize ERP amplitudes, for example, by vector length, prior to testing for interactions. The assumptions of this procedure are examined and it is shown via simulations that this application of vector scaling is both conceptually flawed and unsound in experimental practice. Because different spatial configurations of neural generators cannot reliably be inferred from different scalp topographies even after amplitude normalization, it is recommended that the procedure no longer be used for this purpose.