Immune checkpoint inhibitors (ICPI) have dramatically shaped the treatment for non-small cell lung. We have witnessed durable responses which have led to unprecedented long for some patients, even in case of advanced or metastatic disease. Sadly, despite this incredible benefit the vast majority of the patients does not benefit or not substantially from being exposed to ICPI. Important downsides are the fact that some of the patients experience permanent and significant immunotoxicities. It has been therefore evident the need to better identify the patients who could more likely benefit for such a treatment. Thus biomarkers that predict response to therapy are a key aspect. Many studies have been published on different biomarkers to address this unmet need. For the time being the only one that has been prospectively validated is the tumor programmed death ligand 1(PD-L1) expression. The level of PD-L1 expression is determined by immunohistochemistry. PD-L1 expression has showed a strong correlation with treatment response and led to overall survival benefit. Several limitations about this biomarker exists, hence the need to develop further and more performing ones. The tumor mutational burden has emerged as possible alternative or complementary biomarker, and some studies have showed that knowing the mutation load of the cancer could predict the likelihood of response and benefit from immune checkpoint inhibitors. The importance played by mutations in specific genes has led to explore in depth gene expression signatures. Another biomarker under study is the density of tumor-infiltrating lymphocytes in the tumor microenvironment. Contradicting results have been seen and several studies are ongoing at the moment. Many biomarkers are linked to the presence of cancer tissue to carry further analysis. Several groups have worked on peripheral serum markers as a possible alternative to tissue based one. Another biomarker that remains controversial is the microbiota. Last but no least the application on artificial intelligence in this setting remains quite fascinating, even though no clear data had establish how this could be easily implemented in real practice. In this present work I have reviewed the evidence about the more promising biomarkers in order to clarify the present with a glimpse at the future.