Anti-citrullinated protein antibodies (ACPA) often precede onset of rheumatoid arthritis (RA) by years, and there is an urgent clinical need for predictors of arthritis development among such at-risk patients. This study assesses the prognostic value of ultrasound for arthritis development among ACPA-positive patients with musculoskeletal pain. We prospectively followed 82 ACPA-positive patients without clinical signs of arthritis at baseline. Ultrasound at baseline assessed synovial hypertrophy, inflammatory activity by power Doppler, and erosions in small joints of hands and feet. We applied Cox regression analyses to examine associations with clinical arthritis development during follow-up (median, 69 months; range, 24-90 months). We also compared the ultrasound findings among the patients to a control group of 100 blood donors without musculoskeletal pain. Clinical arthritis developed in 39/82 patients (48%) after a median of 6 months (range, 1-71 months). One or more ultrasound erosions occurred in 13/82 patients (16%), with none in control subjects ( < 0.001). Clinical arthritis development was more common among patients with baseline ultrasound erosions than those without (77 vs. 42%, = 0.032), and remained significant in a multivariable Cox regression analysis that included previously described prognostic factors (HR 3.9, 95% CI 1.6-9.4, = 0.003). Ultrasound-detected tenosynovitis was more frequent among the patients and associated with clinical arthritis development in a univariable analysis (HR 2.5, 95% CI 1.1-5.7, = 0.031), but did not remain statistically significant in multivariable analysis. Thus, bone erosions detected by ultrasound are independent predictors of clinical arthritis development in an ACPA-positive at-risk population. Regional Ethics Committee in Linköping, Sweden, Dnr M220-09. Registered 16 December 2009, https://etikprovningsmyndigheten.se/.