The objective of the study was, with multidetector computed tomography (CT) as the reference method, to determine whether bone erosions in rheumatoid arthritis (RA) metacarpophalangeal (MCP) joints detected with magnetic resonance imaging (MRI) and ultrasonography (US), but not with radiography, represent true erosive changes. We included 17 RA patients with at least one, previously detected, radiographically invisible MCP joint MRI erosion, and four healthy control individuals. They all underwent CT, MRI, US and radiography of the 2nd to 5th MCP joints of one hand on the same day. Each imaging modality was evaluated for the presence of bone erosions in each MCP joint quadrant. In total, 336 quadrants were examined. The sensitivity, specificity and accuracy, respectively, for detecting bone erosions (with CT as the reference method) were 19%, 100% and 81% for radiography; 68%, 96% and 89% for MRI; and 42%, 91% and 80% for US. When the 16 quadrants with radiographic erosions were excluded from the analysis, similar values for MRI (65%, 96% and 90%) and US (30%, 92% and 80%) were obtained. CT and MRI detected at least one erosion in all patients but none in control individuals. US detected at least one erosion in 15 patients, however, erosion-like changes were seen on US in all control individuals. Nine patients had no erosions on radiography. In conclusion, with CT as the reference method, MRI and US exhibited high specificities (96% and 91%, respectively) in detecting bone erosions in RA MCP joints, even in the radiographically non-erosive joints (96% and 92%). The moderate sensitivities indicate that even more erosions than are seen on MRI and, particularly, US are present. Radiography exhibited high specificity (100%) but low sensitivity (19%). The present study strongly indicates that bone erosions, detected with MRI and US in RA patients, represent a loss of calcified tissue with cortical destruction, and therefore can be considered true bone erosions.