In this paper, the direct-projection method given by Benzi and Meyer (1995)
is derived by slightly different way - Basic Solutions of corresponding ho
mogeneous system. The idea of the method differs from the idea of the Gauss
ian Elimination (or LU decomposition). The method works for every nonsingul
ar coefficient matrix in the absence of rounding error. The Gaussian Elimin
ation is explained by the method. The corresponding numerical algorithms of
the method are also given, The method is applied to find the preconditione
r for least squares problems and solvers of singular systems. The numerical
experiments illustrate that the method has better numerical stability than
the Gaussian Elimination.