We propose a subspace based blind identification algorithm using CGM (conjugate gradient method). The algorithm estimates (1) the channel order, (2) the noise variance, (3) the noise subspace, and them identifies (4) channel impulse response from second order statistics of received signal without using the eigenvalue decomposition. The special features of the proposed algorithm are (1) accurate channel order estimation and (2) the reduction of computational complexity using CGM. Numerical examples show the effectiveness of the proposed algorithm.