Abstract: Although the protein monomer structure can be accurately predicted by alphafold2, the structure prediction of multimeric protein complex is still an urgent problem to be solved. Alphafold3 and alphafold multimer attempted to predict the structure of multimeric protein complexes and simulate the interaction between proteins and small molecules, but the results were not satisfactory. One of the main reasons for the above situation is that the interaction interface prediction is not accurate. To solve this problem, we proposed a multimeric protein complex interaction prediction method called PamaMultimer. In our research, we term the aggregation of amino acids in proximity to the central amino acid as a "patch". By considering the solvent accessible surface area and internal contact area of the patch, we give the most likely interface patch and interface amino acids. It is worth mentioning that we used all the multimeric protein complexes (chain number greater than 2, monomers are proteins) that can be searched in the RCSB pdb library for testing, and the success rate and qualification rate of interface patch prediction reached 64.74% and 96.07%. The comprehensive and unomitted dataset strongly supports the accuracy and effectiveness of PamaMultimer. In addition, on the challenging antigen-antibody dataset, PamaMultimer predicted the top 10 binding sites with 48% accuracy. It has obvious advantages in comparison with the state-of-the-art methods. Different from the method of predicting amino acid binding sites, we use the patch as a whole to predict the interface. This helps to identify clusters of amino acids that work together to perform a function, which may help biologists make valuable discoveries. The code and data are available at https://github.com/hyx-1/PamaMultimer.