GSO (Gram-Schmidt Orth normalization) is based on depth function of Euclidean vector which can be proposed to compute the projection depth. The performance of GSO can be studied to exact and approximate algorithms, bivariate data (data from two variables) can be used to associate estimation namely Stahel-Donoho (S-D) location and scatter estimation. The efficiency can be checked by computing average misclassification error in discriminant analysis under real and stimulating environment.