Ultrafast computed tomography (CT) is a non-invasive method of visualizing and quantifying coronary artery calcification; its reproducibility, however, has not been fully elucidated.To assess intra-observer, inter-observer, and inter-study reproducibility, 75 consecutive patients (51 men and 24 women) were studied. CT images were obtained using the volume mode of ultrafast CT (Imatron C-100). A total coronary calcification score (TCS) was calculated from the lesion area (> or = 2 pixels) and its peak CT density (> or = 130 HU).There was no intra-observer variability in two experienced observers. The TCS provided by these observers disagreed in 18 out of 75 (24%) cases, and the differences were -5.1 +/- 53 (mean +/- SD) for TCS and 0.014 +/- 0.13 for In(1 + TCS). They resulted from either 10 incorrect identifications of small coronary branches, or eight variations in determination of the ostial margin. The former was much smaller than the latter in TCS (0.66 +/- 3.0 and -48 +/- 165, respectively), but both were quite similar in In(1 + TCS) (0.082 +/- 0.31, 0.017 +/- 0.22, respectively). Between two scans, 50 out of 75 patients (67%) had different TCS values. The mean differences (95% confidence interval) were 1.8 +/- 106 (-210 to 214) in TCS, and -0.015 +/- 0.46 (-0.94 to 0.91) in In(1 + TCS). Because the differences increased with the mean values, the determination of TCS assumed a constant variance with increasing mean level. A comparison of scan images indicated that partial volume effects were responsible for this constant variance.Partial volume effects play a key role in producing the variability of TCS determination, and log transformation should be used to interpret TCS values. Thus, for clinical purposes, we recommend that two scans be performed in rapid succession, and that the average of these two scans be used to determine TCS.