In this paper, the new computational formulas are derived for the effective condition number Cond eff, and the new error bounds involved in both Cond and Cond eff are developed. A theoretical analysis is provided to support some conclusions in Banoczi et al. (SIAM J. Sci. Comput. 1998; 20:203-227). For the linear algebraic equations solved by the Gaussian elimination or the QR factorization (QR), the direction of the right-hand vector is insignificant for the solution errors, but such a conclusion is invalid for the finite difference method for Poisson's equation. The effective condition number is important to the numerical partial differential equations, because the discretization errors are dominant. EFFECTIVE CONDITION NUMBER FOR NUMERICAL PDE 577 where = A / n <1. In this paper the following new computational formula is derived: Dx x ‡ The exact solution of (5) may not exist. Then, the solutions are regarded as the least-squares solutions as min x∈R n Fx−b . This displays an important role of Cond eff for the stability in the numerical PDE due to the dominant discretization errors. Note that N 32 and then h 1 32 1 145 , Equation (87) coincides with Proposition 4.2 and the analysis made above. † † The conclusions in [2] and this paper are also valid if under a relaxed assumption: S f w k O(1), where w k is the eigenfunction of the leading eigenvalue k of (107).