Dear Sirs, We congratulate the authors on this interesting paper [1]. The authors report that positron emission tomography with glucose analog fluorodeoxyglucose (FDG-PET) had a high accuracy to predict tumour response to neoadjuvant chemotherapy. In patients with adenocarcinoma of the oesophagogastric junction, metabolic changes within the first 2 weeks of therapy were at least as predictive as histopathologic response. It is remarkable, however, that the graphically depicted exit measurements for the standard uptake value (SUV mean) of FDG-PET show a minimal range of variation. The graph presents a mean tumour FDG uptake of 10.7, with a standard deviation of ±1.2 at baseline for responders and 7.1± 0.1 for non-responders (Fig. 1a). Based on this graph, one might consider it possible to predict therapeutic response with the PET examination even prior to initiation of therapy. The authors discuss this and conclude that the difference between exit measurements and the SUV value after 2 weeks allows even better differentiation between responders and non-responders. However, it is also remarkable that PET measurements of advanced tumours in 24 patients should show so little variation. The data assembled in the table do not correspond with those in Fig. 1a. A proper plotting of the data corresponding to the “Patients and methods” section of this paper yields a completely different picture. Fig. 1b depicts the results with the correct values. It is clear that both the exit values and the values after therapy are subject to large variations. Similarly, if the correct graphic presentation is used, the significant difference of the average reduction of SUV mean values between histopathologic responders and nonresponders is no longer evident (Figs. 2a and b). The authors identified a cut-off for SUV values to predict response to neoadjuvant chemotherapy after 2 weeks of therapy. They found an optimum threshold value of −33% to differentiate between minor and major response. The receiver operating characteristic (ROC) curve demonstrated that the highest accuracy for differentiation of histopathologically responding and non-responding tumours was achieved by applying this cut-off value of a 33% decrease of baseline FDG uptake. In a ROC curve, the true positive rate (sensitivity) is plotted as a function of the false positive rate (100 minus specificity) for different cut-off points. Each point on the ROC plot represents a sensitivity/specificity pair corresponding to a particular decision threshold. A test with perfect discrimination (no overlap in the two distributions) has an ROC plot that passes through the upper left corner (100% sensitivity, 100% specificity). Therefore, the closer a ROC plot is to the upper left corner, the higher the overall accuracy of the test [2]. In the current case, the published ROC curve is reproducible from the given data. However, adding 95% confidence intervals to this curve confuses the results (Fig. 3). The 95% confidence interval is the interval in which the true (population) area under the ROC curve lies within 95% confidence. Eur J Nucl Med Mol Imaging (2008) 35:1742–1743 DOI 10.1007/s00259-008-0851-9