Many nuclear medicine physicians use criteria put forth by McNeil or Biello et al. to interpret lung scintigrams, while others use an integration of these and other reported or personal criteria. Interobserver variation within these schemes has not been assessed, nor have the schemes been compared with each other using the same set of scans. Two experienced nuclear medicine physicians independently read 98 scans that were collected over a two year period and that had angiographic correlation. Perfusion, ventilation, and radiographic abnormalities were described objectively, and each case was assigned to a McNeil category, a Biello category, and an estimate of probability was made based on the reader's own synthesis of criteria (modified criteria). The two readers differed widely in describing the number, size, and location of abnormalities. Despite the large interobserver variation in objective descriptions, there were no statistically significant differences between the receiver operating characteristic (ROC) curves for each reader. When the ROC curves for each set of criteria were compared, the results did not confirm that one set of criteria was clearly superior to another. The Biello and modified criteria did classify fewer cases as indeterminate or intermediate probability without a drop in overall accuracy. When data were subdivided by year, analysis still showed no statistically significant differences between criteria or readers. There was, however, a significant difference (P less than .001) in overall accuracy when ROC curves for the two years were compared. This has implications for data obtained in retrospective studies of lung scintigraphy.