Objective The Cohort Hip and Cohort Knee (CHECK) is a cohort of middle‐aged individuals with hip or knee pain. Radiographs were assigned Kellgren/Lawrence (K/L) scores under different conditions at each follow‐up visit for 10 years. We aimed to describe and consolidate each scoring approach, then illustrate implications of their use by comparing baseline K/L scores assigned using 2 of these approaches, and evaluating their respective associations with joint replacement and incident radiographic osteoarthritis (ROA). Methods We compared baseline K/L scores assigned to hips and knees using 2 scoring approaches: 1) assigned by senior researchers to baseline images alone and 2) assigned by trained readers, with images read paired and in known sequence with up to 10 years of follow‐up radiographs (Poisson regression). We evaluated the associations of baseline ROA (any: K/L grade ≥1; established: K/L ≥2) with joint replacement, and of K/L 1 joints with incident established ROA (survival analysis). Results Of 1,002 participants (79% women, mean ± SD age 55.9 ± 5.2 years, body mass index 26.2 ± 4.0 kg/m 2 ), the second scoring approach had 2.4 times (95% confidence interval [95% CI] 1.8–3.1 for knees) and 2.9 times (95% CI 2.3–3.7 for hips) higher prevalence of established ROA than the first approach. Established hip ROA had a higher risk of joint replacement using the first approach (hazard ratio [HR] 24.2 [95% CI 15.0–39.8] versus second approach HR 7.7 [95% CI 4.9–12.1]), as did knees (HR 19.3 [95% CI 10.3–36.1] versus second approach HR 4.8 [95% CI 2.4–9.6]). The risk of incident ROA did not differ by approach. Conclusion This study demonstrates that evaluating ROA prevalence and predicting outcomes depends on the scoring approach.