To estimate and compare the diagnostic accuracy of magnetic resonance imaging (MRI) and ultrasound, for the prediction of rheumatoid arthritis (RA) in unclassified arthritis (UA)./r/nMEDLINE, Embase and BIOSIS were searched from 1987 to May 2019. Studies evaluating any imaging test in participants with UA were eligible. Reference standards were RA classification criteria or methotrexate initiation. Two authors independently extracted data and assessed validity using QUADAS-2. Sensitivities and specificities were calculated for each imaging characteristic and joint area. Summary estimates with 95% confidence intervals (CI) were estimated where possible./r/nNineteen studies were included; 13 evaluated MRI (n=1,143; 454 with RA) and 6 evaluated ultrasound (n=531; 205 with RA). Studies were limited by unclear recruitment procedures, inclusion of patients with RA at baseline, differential verification, lack of blinding and consensus grading. Study heterogeneity largely precluded meta-analysis, however summary sensitivity and specificity for MRI synovitis in at least one joint were 93% (95% CI 88%, 96%) and 25% (95% CI 13%, 41%) (3 studies). Specificities may be higher for other MRI characteristics but data are limited. Ultrasound results were difficult to synthesise due to different diagnostic thresholds and reference standards./r/nThe evidence for MRI or ultrasound as single tests for predicting RA in people with UA is heterogeneous and of variable methodological quality. Larger studies using consensus grading and consistently defined RA diagnosis are needed to identify whether combinations of imaging characteristics, either alone or in combination with other clinical findings, can better predict RA in this population.