Objective: Epidemiological studies of knee osteoarthritis (OA) have often used a radiographic definition. However, the clinical syndrome of OA is influenced by a broad range of factors in addition to the structural changes required for radiographic OA. Hence more recently several studies have adopted a clinical or self-reported approach to OA diagnosis rather than a radiographic approach. The aim of this study was to investigate agreement between radiographic OA and the clinical and self-reported diagnoses of OA. Design: Data were available for 199 men and 196 women in the Hertfordshire Cohort Study (HCS), UK. Participants completed a questionnaire detailing self-reported OA. Clinical OA was defined based on American College of Rheumatology (ACR) criteria. Knee radiographs were taken and graded for overall Kellgren and Lawrence (K&L) score. Results: The mean (standard deviation (SD)) age of study participants was 75.2 (2.6) years and almost identical proportions of men and women. The prevalence of knee OA differed depending on the method employed for diagnosis; 21% of the study participants self-reported knee OA, 18% of the participants had clinical knee OA and 42% of the participants had radiographic OA. Of those 72 study participants with a self-reported diagnosis of knee OA 52 (72%) had a radiographic diagnosis of knee OA, while 66% (39 out of 59) of study participants with clinical knee OA had a diagnosis of radiographic knee OA. However 58% of those participants diagnosed with radiographic OA did not have either self-reported knee OA or a diagnosis of clinical OA. Therefore in comparison with the radiographic definition of OA, both the clinical and self-report definitions had high specificity (91.5% & 91.5% respectively) and low sensitivity (24.5% and 32.7% respectively). Conclusion: There is modest agreement between the radiographic, clinical and self-report methods of diagnosis of knee OA.