BACKGROUND: Evidence-based models are needed to deliver exercise-related services for knee osteoarthritis efficiently and according to patient needs. OBJECTIVE: To examine a stepped exercise program for patients with knee osteoarthritis (STEP-KOA). DESIGN: Randomized controlled trial. (ClinicalTrials.gov: NCT02653768). SETTING: 2 U.S. Department of Veterans Affairs sites. PARTICIPANTS: 345 patients (mean age, 60 years; 15% female; 67% people of color) with symptomatic knee osteoarthritis. INTERVENTION: Participants were randomly assigned in a 2:1 ratio to STEP-KOA or an arthritis education (AE) control group, respectively. The STEP-KOA intervention began with 3 months of an internet-based exercise program (step 1). Participants who did not meet response criteria for improvement in pain and function after step 1 progressed to step 2, which involved 3 months of biweekly physical activity coaching calls. Participants who did not meet response criteria after step 2 went on to in-person physical therapy visits (step 3). The AE group received educational materials via mail every 2 weeks. MEASUREMENTS: Primary outcome was Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) score. Scores for the STEP-KOA and AE groups at 9 months were compared by using linear mixed models. RESULTS: In the STEP-KOA group, 65% of participants (150 of 230) progressed to step 2 and 35% (81 of 230) to step 3. The estimated baseline WOMAC score for the full sample was 47.5 (95% CI, 45.7 to 49.2). At 9-month follow-up, the estimated mean WOMAC score was 6.8 points (CI, -10.5 to -3.2 points) lower in the STEP-KOA than the AE group, indicating greater improvement. LIMITATION: Participants were mostly male veterans, and follow-up was limited. CONCLUSION: Veterans in STEP-KOA reported modest improvements in knee osteoarthritis symptoms compared with the control group. The STEP-KOA strategy may be efficient for delivering exercise therapies for knee osteoarthritis. PRIMARY FUNDING SOURCE: Department of Veterans Affairs, Health Services Research and Development Service.