The importance of controlling grain boundary (GB) segregation is increasing with the strengthening of steel. In this study, a theoretical prediction method for the amount of GB segregation for a solute element in polycrystals, is established. In this prediction method, a nano-polycrystalline GB model for simulating GBs in polycrystals is developed, and the segregation energy of a solute element is calculated comprehensively for all atomic sites constituting the GB model by using an interatomic potential. From the obtained segregation energies, the segregation amount of the solute element at each atomic site is determined. Subsequently, each atomic site is classified for based on its distance from the GB center, and averaged to calculate the segregation profile of the solute element for that distance from the GB center. By applying this method to the GB segregation of P in bcc-Fe and comparing its results with experimental findings, it is determined that this prediction method can attain a good prediction accuracy.