In cancer immunotherapy, the precise and efficient identification to HLA-associated peptides (HAP) is a critical step, while liquid chromatography-tandem mass spectrometry (LC-MS/MS) is a premising means to discover these HAPs that are partially not derived from the traditional annotation of the human genome. The HAPs are generally considered as the endogenous peptides cleaved by uncertain enzyme(s). In contrast to a conventional search upon tryptic peptides, the non-specific cleavage peptides are expected with their unique features and are identified through a search specially optimized. By inspecting several publicly available datasets and home-made ones, we found the search results for HAPs quite unstable with lower identification rate and high false positive rate by using several common search engines, such as Mascot, Comet, X! Tandem and MSGF+ and some post-processor, like PeptideProphet and Percolator. Herein, we have focused on the characteristics of non-specific cleavage peptides and optimized the post-search process to improve the estimation of posterior error rate. In the primary test using Mascot, with only one additional feature and percolator training, the identified PSMs in two datasets increased from 1,401 to 1,986 and from 9,133 to 9,351, respectively. With more features added, the results were further improved. In future, we are looking forward to finding a proper combination of features and training strategy to further optimize the post-process with a robust increase of identification rate and consistency using different search engines. And the optimization is expected not only limited in the identification of HAPs but also beneficial to the identification of all the peptidome with non-specific cleavage search involved.