In this paper, we present a method based on mining maximal frequent patterns for core-attachment complexes identification in yeast protein-protein interaction networks (PINs). Our method contains of two stages. Firstly, it finds all the protein-complex cores by mining maximal frequent patterns in PIN using FP-growth method. Then it filters the redundant cores and adds the attachment proteins for each remained core to form protein complexes. We experimentally evaluate the performance of our method using three different yeast PINs. The results show that our method is better than other existing methods with regard to localization and Gene Ontology (GO) semantic similarity within the predicted complexes. Furthermore, the accuracy of prediction with regard to the known CYC2008 reference complexes proves that our results can obtain higher map complex rate.