Despite few and inconclusive studies, radiography is generally believed to be insensitive for detection of osteolytic lesions of the spine. A more detailed investigation was undertaken to study the detectability of laboratory-produced osteolytic lesions in cadaveric thoracolumbar vertebral bodies using conventional lateral radiographs. The radiographs were presented to four radiologists in two sessions over a two month period. In the first session, the films were arranged in a composite of five vertebral bodies, T11 to L3 all from the same spine, in which one contained a lesion and the other four were normal. In the second session, each vertebral body film was presented individually. Area (Az) under the receiver-operating characteristic (ROC) curve was used to measure the performances of readers. Observer detection was similar in the two formats with Az ranging from 0.67 +/- 0.05 to 0.79 +/- 0.04 for the composite film arrangement and 0.57 +/- 0.08 to 0.85 +/- 0.10 for the films of individual vertebral bodies. Lesions were grouped into three relative size categories: 18% to 25%, 26% to 40%, and 41% to 60% of transverse vertebral body diameter. The mean increase in ROC area between the small and large lesions was 0.29 (P less than 0.04) for the composite films and 0.16 (P less than 0.05) for the individual films. In the composite study, all readers showed significant (P less than 0.05) increases in lesion detection in spines reflecting large increases (P less than 0.01) in bone mineral content.(ABSTRACT TRUNCATED AT 250 WORDS)