Purpose/Objective: In daily practice, several different delineation guidelines can be used for the delineation of organs at risk that are involved in swallowing (SWOARs). Based on these delineation guidelines, predictive models have been developed for the different aspects of post treatment swallowing dysfunction. This study was performed to test the hypothesis that different guidelines result in different delineated SWOARs and subsequently different NTCP values for swallowing dysfunction. Materials and Methods: Thirty head and neck cancer patients with either laryngeal or pharyngeal carcinoma, treated with swallowing sparing IMRT, were included in this study. Nine SWOAR delineation guidelines derived from literature were compared. The following SWOARs were investigated: superior pharyngeal constrictor muscles (PCM), middle PCM, supraglottic larynx and esophageal inlet muscle (EIM). Especially, differences in description, geometric measures (centre of mass and coefficients of variance for volume), and resulting SWOAR dose-volume parameters were investigated. The results were used to examine the variation in the prediction of normal tissue complication probability (NTCP). For this comparison, five NTCP models for swallowing dysfunction from literature were used with the following endpoints: (1) RTOG grade 2-4; (2) problems with swallowing solid, (3) soft, and (4) liquid food; (5) choking when swallowing. Results: The description of the posterior and anterior borders of the SWOARs in the delineation atlases was similar. However, differences up to several centimeters were seen in the description of the cranial and caudal borders of some of the SWOARs. Mean differences in the position of the centre of mass in cranio-caudal direction were largest for the EIM and superior PCM: 2.2 cm ± 0.4 cm and -1.2 cm ± 0.6 cm respectively. Coefficients of variance of the volume of the SWOARs were largest for the EIM and for the middle PCM: 80% and 45% respectively. The range in differences in NTCP value per tumor location for all models was 0.1 - 18% for oropharynx (18% for endpoint 2), 0.00 - 22.4% for larynx (22.4% for endpoint 5), 0.04 - 7.7% for nasopharynx (7.7% for endpoint 5), 0.20 - 11.7% for hypopharynx (11.7% for endpoint 5), and 0.04 - 1.7% for oral cavity (1.7% for endpoint 2). Thus largest variations were found for tumors situated in the larynx and for NTCP model 'Choking when swallowing' (see figure). (Figure presented). Conclusions: Large variations in the definition of SWOARs were found among published delineation guidelines. This caused considerable variation in delineated volumes and in the position of the SWOARs which in turn resulted in different dose-volume values, and in different NTCP values. Therefore, an NTCP model from a specific group should only be used when SWOARs are delineated according to the corresponding delineation guidelines. Moreover, formulation of generally accepted guidelines for swallowing structures is strongly recommended to improve the generalisability of NTCP models.