INTRODUCTION: Sampling is regarded as crucial to the validity of case-control studies. Ideally, cases and controls should be selected from the same source population, but deviations from this approach are often seen. OBJECTIVE: Our objective was to examine how exposure-disease relationships in a study on chronic obstructive pulmonary disease (COPD) were affected by the sampling sources of cases and controls. METHODS: A Norwegian case-control study on COPD including 1909 subjects used three sources of recruitment for cases (general population, hospital registry and volunteers) and two sources for controls (general population and volunteers). This resulted in six sampling combinations of cases and controls (groups A-F). We examined how the risk factors gender, age, smoking, educational level and comorbidity were associated with COPD in these six sampling groups. RESULTS: Several exposure-disease associations were dependent on variation in sampling source, thereby demonstrating the possibility of selection bias. The theoretically most ideal sampling group is likely group A, where both cases and controls are recruited from a general population. When using group A as a reference, the groups containing either voluntary controls and/or hospital-based cases deviated the most, suggesting higher susceptibility to selection bias in these groups. CONCLUSION: Recruitment from several sources made our study design vulnerable to selection bias. Our findings should bring about increased awareness to the sampling process, and encourage sampling of cases and controls from the same source population in future studies.