The invention discloses a protein conformation space search method guided by double distribution estimation. Multiple Monte Carlo tracks are enabled at the same time; an energy distribution probability model and a historical acceptance probability model are built according to information of current all track individuals; one Monte Carlo track is selected according to the two probability models and the next search is executed, so that a search process is mainly carried out in a region with lower energy; a search is performed in a high-energy region according to a certain probability, so that the problem of inaccurate energy model is solved to a certain extent; and finally a series of near natural conformations are obtained. The method is applied to the protein structure prediction; and the conformations with relatively high prediction precision and relatively low complexity can be obtained.