A parametric approach to measure the galaxy morphological structure suffers from systematic and random errors owing to the characteristics of chosen model and data which are difficult to include in the inference of galaxy morphology distribution. In this talk, I will introduce a Bayesian MCMC approach, fully considering the posterior probability for parameter estimation and model inference. Important issues in galaxy image decomposition, which are often neglected due to the limitations of conventional analysis methods, will be highlighted and significant improvements in parameter estimation and model inference will be shown by using an ensemble of simulated and real 2MASS galaxy samples.