Introduction: We compared the performance of real-time PCR with culture-based methods for identifying bacteria in sputum samples from patients with chronic obstructive pulmonary disease (COPD) in three studies. Methods: This was an exploratory analysis of sputum samples collected during an observational study of 127 patients (AERIS; NCT01360398), phase 2 study of 145 patients (NTHI-004; NCT02075541), and phase 2b study of 606 patients (NTHI-MCAT-002; NCT03281876). Bacteria were identified by culture-based microbiological methods in local laboratories using fresh samples or by real-time PCR in a central laboratory using frozen samples. Haemophilus influenzae positivity with culture was differentiated from H. haemolyticus positivity by microarray analysis or PCR. The feasibility of bacterial detection by culture-based methods on previously frozen samples was also examined in the NTHI-004 study. Results: Bacterial detection results from both culture-based and PCR assays were available from 2,293 samples from AERIS, 974 from the NTHI-004 study, and 1736 from the NTHI-MCAT-002 study. Quantitative real-time PCR (qPCR) showed higher positivity rates than culture for H. influenzae (percentages for each study: 43.4% versus 26.2%, 47.1% versus 23.6%, 32.7% versus 10.4%) and Moraxella catarrhalis (12.9% versus 6.3%, 19.0% versus 6.0%, 15.5% versus 4.1%). In the NTHI-004 and NTHI-MCAT-002 studies, positivity rates were higher with qPCR for Streptococcus pneumoniae (15.6% versus 6.1%, 15.5% versus 3.8%); in AERIS, a lower rate with qPCR than with culture (11.0% versus 17.4%) was explained by misidentification of S. pseudopneumoniae/mitis isolates via conventional microbiological methods. Concordance analysis showed lowest overall agreement for H. influenzae (82.0%, 75.6%, 77.6%), due mainly to culture-negative/qPCR-positive samples, indicating lower sensitivity of the culture-based methods. The lowest positive agreement (culture-positive/qPCR-positive samples) was observed for S. pneumoniae (35.1%, 71.2%, 71.2%). Bacterial load values for each species showed a proportion of culture-negative samples with a load detected by qPCR; for some samples, the loads were in line with those observed in culture-positive samples. In the NTHI-004 study, of fresh samples that tested culture-positive, less than 50% remained culture-positive when tested from freeze/thawed samples. In the NTHI-004 study, of fresh samples that tested culture-positive, less than 50% remained culture-positive when tested from freeze/thawed samples. Discussion: Real-time PCR on frozen sputum samples has enhanced sensitivity and specificity over culture-based methods, supporting its use for the identification of common respiratory bacterial species in patients with COPD.