The representation of test scores as n dimensional points leads directly to an estimate of error variance at a particular score level in the case of equivalent items. Approximations are suggested for the case of non-equivalent items. These approximations are compared with empirical data prepared by Dr. Mollenkopf with satisfactory results.