Chronic obstructive pulmonary disease (COPD) has a profound impact on daily life, yet remains underdiagnosed and undertreated. This study aims to discover potential protein biomarkers for diagnosis and classification of COPD. Fifty-seven COPD patients and 40 controls were divided into a training set (30 COPD patients, 20 healthy controls) and a test set (27 COPD patients, 20 healthy controls). Serum proteomic profiles were measured using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). A classification tree was established using Biomarker Pattern Software (BPS). Next we screened distinct proteins present in patients with acute exacerbations of COPD (AECOPD), stable COPD and healthy controls, in order to establish diagnostic models for classification of COPD. Twenty peaks showed statistically significant differences between COPD patients and healthy controls (P < 0.05). Two proteomic peaks (3167 and 5477 m/z) were chosen by BPS to establish a classification tree in the training set. The sensitivity and specificity of this classification tree were 92.59% and 90.00% respectively in the testing set. Furthermore, differently expressed proteins were detected among the patients with AECOPD, stable COPD, and healthy controls. Two protein profiles (3167 and 4645 m/z) could distinguish between stable COPD patients and healthy controls. Three protein profiles (3167, 2963 and 2973 m/z) could distinguish between AECOPD patients and healthy controls. Three protein profiles (5476, 14039 and 2831 m/z) could distinguish between stable COPD patients and AECOPD patients. SELDI-TOFMS Proteinchip technology is a quick, easy and practical, high throughput analytic method. It shows the diagnostic models established by distinguished proteomic peaks can discriminate COPD patients from healthy control and can identify different stages of COPD. It will provide a highly accurate approach for diagnosis and clinical staging of COPD.