was first obtained by Subbotin (see Subbotin [15] and Johnson and Kotz [5]). This distribution exhibits many remarkable properties at various points as seen in the works of Box and Tiao [1], Diananda [2] and Johnson and Kotz, op. cit. In this paper, the author tries to characterize this distribution from some other points of view. Namely, the norm is introduced to random variables and density functions (see Rice and White [12]) and the proper order of distribution is defined. The generalized median and dispersion (see Fechner [4]) and some related notions (see Taguchi [16] and [17]) are applied to GED. The estimation of parameters are also discussed. Moreover, the relationships held between GED and some other distributions are examined. On the other hand, the relationships held among n independent random variables having the same or different GED's are exhibited, too. In Section 3, the bivariate general error distributions in the wider sense is also defined and some of them are given in an explicit form through the introduction of new parameters p and 0, which call respectively dependency and asymmetry parameters. In this case, the traditional bivariate Gaussian distribution becomes the special case of this distribution. In the same way, the asymmetric bivariate GED in the wider sense is introduced and an example is given. At the same time, the bivariate GED (in the narrow sense) is defined and one of them is given in the case of p>__l. The standard form of this distribution has only one parameter p p ( l < p p < l ) expressing the correlation coefficient of proper order. If p=2, this distribution is exactly the bivariate Gaussian distribution itself. Generally the bivariate GED in the wider sense is expressible as the mixture of the bivariate GED's having re-