Abstract: In this paper，based on various methods of using extrinsic information, two logarithmic iterative decoding algorithms of Turbo codes are anatomized clearly for a discrete memoryless Gaussian channel. With Monte Carlo simulation, performance of two algorithms is compared under the same constraint condition; after analysis it is proposed that they are identical in nature, the only difference is how to deal with scale of extrinsic information for component decoder to use. Finally, a united form is used to describe two algorithms and some research is done for the dependency on extrinsic information in iterative decoding numerically, from the result it is found that SNR and iterative number do not affect the best scale of extrinsic information used in component decoder.