Seismic inversion to acoustic impedance is widely used in exploration seismology. In recent years, stochastic seismic inversion has been developed significantly. In this paper, a global stochastic inversionmethod is presented in stationary and non-stationary conditions. Application of the methods to a synthetic dataset shows considerable improvement in spatial distribution of acoustic impedance values in non-stationary method. Validation based on two blind well tests also confirms this improvement according to decreasing RMS error in case of non-stationary condition compared to stationary one. In order to evaluate the methods in the presence of noise, a noisy seismic dataset with signal to noise ratio of 4 db has been used.