Somatostatin-receptor (SSTR) scintigraphy using the single photon emission computed tomography (SPECT) technique allows the assessment of orbital inflammation in patients with Graves' disease. Previous studies showed differences in orbital octreotide uptake already 4 hr after injection. In this study, analysis of inter-/intra-observer variance and reproducibility in the evaluation of orbital SPECT images was performed. First, SPECT data of one representative female patient with clinically active Graves' ophthalmopathy (GO), obtained 4 hr after intravenous injection of 110 MBq 111In-pentetreotide and processed by filtered backprojection, were analyzed. Transverse SPECT images were reconstructed, an optimal orbital image was selected and predetermined regions of interests (ROIs) for both orbits were positioned by three independent observers 15 to 19 times each. In a second step, SPECT data of 8 different patients with GO were evaluated in the same manner by four independent observers 3 to 4 times each. Variance component partitioning was used to compare the order of intra- and inter-observer variation. For the right and the left orbit, the inter-observer variance proportion was 90% and 79%, whereas intra-observer variance partition was 10% and 21%, respectively. The corresponding ratios 0.11 and 0.27 summarize the comparison of sources of variance. The overall reliability was 84%, representing the patients influence on the total variance. Intra-observer reliability for both orbits was 88%, 89%, 97% and 98% (mean over orbits), respectively for observers I to IV. Using the Spearman Brown prophecy formula it follows that two replications per patient are sufficient to ensure a minimum reproducibility of 90%, which is also confirmed by the low intra-observer variation. Furthermore, intra-class correlation as a measure of (multiple) observer reproducibility was 94%. In conclusion, due to the increased inter-observer variance proportion and the high variation in intra-observer reliability, evaluations of orbital SSTR scintigraphy have to be done by the same and experienced observer leading to comparable data. But an automatic and quantitative computerized technique for evaluation of these SPECT data should be exactly reproducible and probably lead to more accurate and representative results.