There is growing interest in monitoring response to therapy in oncology patients using positron emission tomography and fluorine-18 fluorodeoxyglucose (FDG). Quantification is required, and various methods have been described. Arterial sampling, which provides the most accurate input function, often is not feasible in patients undergoing chemotherapy. In the thorax an image-derived input function can be obtained from the large vascular structures. In many studies the left ventricle (LV) is used, but the obtained data are rarely validated. In this study a simple quality control procedure for the image-derived input function was developed and compared with a standard LV curve. Twenty dynamic FDG scans were obtained in nine patients with non-small cell lung cancer stage IIIA-N2. Three venous blood samples were taken as a quality control for the image-derived input function. Regions of interest (ROIs) were defined for aorta, LV and left atrium. Input curves were generated according the the standard use of the LV curve and by applying the quality control method to all vascular ROIs. Tumour ROIs were defined and both input functions were used to calculate tumour glucose metabolism (MRglu), using both standard non-linear regression and Patlak analyses. Mean differences in MRglu using "standard" LV and "quality control approved" input functions were 16.2% and 17.5% for non-linear regression and Patlak analyses, respectively, indicating that the use of "standard" LV curve might lead to significant errors. It is concluded that care should be taken in using image derived input functions without appropriate quality control. The proposed procedure is simple and results in significantly more accurate MRglu data.