Immune checkpoint blockade targeting the PD-1/PD-L1 axis has recently demonstrated efficacy and promise in cancer treatment. Appropriate biomarker selection is therefore essential for improving treatment efficacy. However, the establishment of PD-L1 assay in pathology laboratories is complicated by the presence of multiple testing platforms using different scoring systems. Here we assessed the PD-L1 expression in 713 consecutive non-small cell lung carcinomas by four commercially available PD-L1 immunohistochemical assays, namely, 22C3, 28-8, SP142 and SP263. The analytical performances of the four assays and diagnostic performances across clinically relevant cutoffs were evaluated. The prevalence of PD-L1 (22C3) expression was 21% with a ≥50% cutoff and 56% with a ≥1% cutoff. High PD-L1 expression (using a ≥50% cutoff) was significantly associated with male sex (P = 0.001), ever smoking history (P < 0.001), squamous cell carcinoma (P = 0.001), large cell carcinoma (P < 0.001), lymphoepithelioma-like carcinoma (P = 0.006), sarcomatoid carcinoma (P < 0.001), mutant KRAS (P = 0.005) and wild-type EGFR (P = 0.003). Elevated PD-L1 expression was also significantly associated with shorter survival in patients with adenocarcinoma (log-rank P = 0.026) and remained an independent prognostic factor by multivariable analysis. Among the four assays, 22C3, 28-8 and SP263 were highly concordant for tumor cell scoring. With a cutoff of ≥50% (i.e., the threshold for first-line patient selection), inter-rater agreement was high among the three assays with percentage agreement >97%. In conclusion, three PD-L1 assays showed good analytical performance and a high agreement with each other, but not all cases were correctly classified using the same clinical cutoff. Further studies comparing the predictive value of these assays are required to address the interchangeability of these assays for clinical use.