Abstract Background The aim was to determine the accuracy of high-resolution ultrasonography (US) for detecting erosion in the metacarpophalangeal (MCP) and wrist joints of patients with different subtypes of systemic lupus erythematosus (SLE) arthritis, using computed tomography (CT) as the gold-standard reference method. Method The ulnar head, radiocarpal and second to fifth MCP joints in 26 patients with SLE - 9 classified as having rhupus syndrome, 10 as having Jaccoud’s arthropathy (JA) and 7 as having non-deforming non-erosive (NDNE) arthritis - were subdivided into areas and bilaterally evaluated for the presence of bone erosion by CT and US. On CT, erosion volume was scored according to the outcome measures in rheumatology-rheumatoid arthritis magnetic resonance imaging (OMERACT-RAMRIS) score. On US, erosions were semi-quantitatively scored 0–3 according to scoring by ultrasound structural erosion (ScUSSe) systems. Results Erosions were detected by CT in 92/728 areas (12.6 %) and by US in 43/728 areas (5.9 %). Sensitivity, specificity and accuracy of US overall was 36 %, 98 % and 90 % compared with 57 %, 98 % and 93 % in the dorsal and lateral aspects of the second and fifth MCP, which were identified as areas with the best US reliability. Adding wrist joints would capture a larger number of erosions without affecting the accuracy. US detected 90.0 % of CT erosions with bone volume loss >20 % and 51.2 % of erosions with bone volume loss >10 %. Patients with rhupus had a greater number of larger erosions than those with JA or NDNE arthritis, with prevalent involvement of the MCP joints. Overall reliability of US in detecting bone erosions was moderate for rhupus syndrome (0.55) and JA (0.58), but poor for NDNE arthritis (0.10). Conclusion US had moderate sensitivity and excellent specificity for detection and semi-quantitative assessment of bone erosions in SLE.