Abstract Objective: To describe the performance of a non-fluoroscopic fixed-flexion PA radiographic protocol with a new positioning device, developed for the assessment of knee osteoarthritis (OA) in Brazilian Longitudinal Study of Adult Health Musculoskeletal Study (ELSA-Brasil MSK). Material and methods: A test–retest design including 19 adults (38 knee images) was conducted. Feasibility of the radiographic protocol was assessed by image quality parameters and presence of radioanatomic alignment according to intermargin distance (IMD) values. Repeatability was assessed for IMD and joint space width (JSW) measured at three different locations. Results: Approximately 90% of knee images presented excellent quality. Frequencies of nearly perfect radioanatomic alignment (IMD ≤1 mm) ranged from 29% to 50%, and satisfactory alignment was found in up to 71% and 76% of the images (IMD ≤1.5 mm and ≤1.7 mm, respectively). Repeatability analyses yielded the following results: IMD [SD of mean difference = 1.08; coefficient of variation (%CV) = 54.68%; intraclass correlation coefficient (ICC) (95%CI) = 0.59 (0.34–0.77)]; JSW [SD of mean difference = 0.34–0.61; %CV = 4.48%–9.80%; ICC (95%CI) = 0.74 (0.55–0.85)–0.94 (0.87–0.97)]. Adequately reproducible measurements of IMD and JSW were found in 68% and 87% of the images, respectively. Conclusions: Despite the difficulty in achieving consistent radioanatomic alignment between subsequent radiographs in terms of IMD, the protocol produced highly repeatable JSW measurements when these were taken at midpoint and 10 mm from the medial extremity of the medial tibial plateau. Therefore, measurements of JSW at these locations can be considered adequate for the assessment of knee OA in ELSA-Brasil MSK.