The shotgun strategy applying tandem mass spectrometry to identify proteins has been widely used in proteomics for its high reliability and efficiency. However, protein deduction is ambiguous due to uncorrected identified peptides and erased correlated information between peptides and their source proteins. Protein quality control methods can be divided into two categories: protein assembly and protein confidence evaluation. To date, parsimony principle, deriving the minimal proteins accounting for all identified peptides, has been widely used in protein assembly. Boolean and probability assembly are two kinds of methods in protein assembly. Protein confidence evaluation includes protein probability calculation of a single protein and false discovery rate estimation of all identified proteins. This review reveals the trend of developing a generalized probabilistic model with consideration of all influence factors, which can be applicable to a variety of peptide scoring system for protein identification.