Objective To determine the incidence and prevalence of knee osteoarthritis (OA) using codified and narrative data from general practices throughout The Netherlands. Methods This retrospective cohort study was conducted using the Integrated Primary Care Information database. Patients with codified knee OA were selected, and an algorithm was developed to identify patients with narratively diagnosed knee OA only. Point prevalence proportions and incidence rates among people age ≥30 years were assessed from 2008 to 2019. The association of comorbidities with codified knee OA was analyzed using multivariable logistic regression. Results The positive predicted value of narratively diagnosed knee OA only was 94.0% (95% confidence interval 95% CI 87.4–100%) and for codified knee OA 96.0% (95% CI 90.6–100%). Including narrative data in addition to codified data resulted in a prevalence 1.83–2.01 times higher (over the study years); prevalence increased from 5.8% to 11.8% between 2008 and 2019. The incidence rate was 1.93–2.28 times higher and increased from 9.98 per 1,000 person‐years to 13.8 per 1,000 person‐years between 2008 and 2019. Among patients with codified knee OA, 39.4% were previously diagnosed narratively with knee OA, on average ~3 years earlier. Comorbidities influenced the likelihood of being recorded with codified knee OA. Conclusion Our study of a Dutch primary care database showed that current incidence and prevalence estimates based on codified data alone from electronic health records are underestimated. Narrative data can be incorporated in addition to codified data to identify knee OA patients more accurately.