Conventional radiography in the lateral projection was used to evaluate 25 osteolytic lesions involving vertebral bodies of the thoracolumbar spine. Destruction was calculated as percentages of the maximum transverse diameter and volume of the vertebral body as measured by computed tomography (CT) and the effect of size of lesion on detection was evaluated. Areas (Az) under the receiver-operating characteristic (ROC) curves and the significance of differences were determined from the observations of four radiologists. The mean difference between areas under the ROC curve for lesions involving 32% to 60% of the transverse diameter and for larger lesions of 61% to 93% was 0.12 and significant (P less than 0.05). The effect of the presence of localized loss of vertebral body density, sclerotic bone tissue surrounding the lesion and cortical destruction was evaluated in a second session, in films with lesions the observers scored 4 (probably abnormal) or 5 (definitely abnormal). Cortical destruction was reported to be helpful for detection in 35% of lesions, sclerotic bone surrounding the lesions was helpful in 30%, and a qualitative local decrease in bone density was noted in all lesions. In comparison with the results obtained for the same four observers with experimentally produced lesions in our previous cadaveric study, the clinical lesions of comparable size were not as readily detected. The ROC area for the largest group of clinical lesions (61% to 93%, Az = 0.83 +/- 0.07) was not significantly different from that for a group of smaller cadaveric lesions (41% to 60%, Az = 0.83 +/- 0.05). The smaller clinical lesions (32% to 60%, Az = 0.71 +/- 0.08) were similar in detectability to the experimental lesions of relative diameter 26% to 40% (Az = 0.74 +/- 0.15). Caution should be exercised in the interpretation of conventional radiographs in the lateral projection if there is suspicion of vertebral body destruction.