Properties of a G (``general'') estimate of square loss for discriminant analysis in problems with two normal distributions are studied. If Kolmogorov's asymptotic property holds then the G-estimate is asymptotically normal.