The association between exacerbation aetiology and exacerbation frequency is poorly understood. We analysed 2-year follow-up data from a prospective observational study of patients with chronic obstructive pulmonary disease (COPD) (www.clinicaltrials.gov identifier number NCT01360398) to evaluate year-to-year variation in exacerbation frequency and related aetiology. A total of 127 patients underwent blood and sputum sampling monthly and at exacerbation to detect respiratory infections and eosinophilic inflammation; 103 continued into year 2 and 88 completed both years. The most common bacterial species at stable state and exacerbation was Haemophilus influenzae. Among infrequent exacerbators (one exacerbation per year), the incidence of viral infection at exacerbation was high (60.0% (95% CI 35.1-81.7%) in year 1 and 78.6% (53.4-94.2%) in year 2). Those with more frequent exacerbations tended to have higher relative incidence of bacterial than viral infection. Patients with at least two additional exacerbations in year 2 versus year 1 had a higher risk of H. influenzae colonisation at stable state than those with at least two fewer exacerbations, as detected by culture (OR 1.43 (95% CI 0.71-2.91) versus 0.63 (0.40-1.01), p=0.06) and PCR (1.76 (95% CI 0.88-3.51) versus 0.56 (0.37-0.86), p<0.01). This was not seen with other infection types or eosinophilic inflammation. Analysis of the same cohort over 2 years showed, for the first time, that changes in yearly COPD exacerbation rate may be associated with variations in H. influenzae colonisation.