Predicting response to checkpoint blockade therapy for lung cancer has largely focused on measuring programmed death-ligand 1 (PD-L1) expression on tumor cells. PD-L1 expression is geographically heterogeneous within many tumors, however, and we questioned whether small tissue samples, such as biopsies, might be sufficiently representative of PD-L1 expression for evaluating this marker in lung cancer tumors. To evaluate the extent of variability of PD-L1 expression in small tissue samples, and how that variability affects accuracy of overall assessment of PD-L1 in lung cancer, we scored immunohistochemical staining for PD-L1 in tissue microarray cores from a series of 79 squamous cell lung cancers and 71 pulmonary adenocarcinomas. Our study found substantial inconsistencies for the percentages of cells staining positive for PD-L1 among different tissue microarray cores in many cases of both adenocarcinoma and squamous cell carcinoma. This variable scoring was seen at both high levels and low levels of PD-L1 expression, and by further evaluation of cases with discordant results on full-face sections to assess geographic distribution of staining, we found that discordant results among different tissue microarray cores reflected geographic variation of PD-L1 expression in those tumors. Moreover, we found that as a result of heterogeneous expression, the sensitivity of a single small tissue sample can be as low as 85% for detecting PD-L1 expression at scoring thresholds commonly used in clinical practice. Based on these studies, we conclude that many cases of lung cancer could be inaccurately or variably scored for PD-L1 expression with a single biopsy sample. Accordingly, lung cancer patients can be inconsistently classified for PD-L1 expression status, particularly when a threshold for the percentage of positive cells is used to determine eligibility for checkpoint blockade therapy.