We have compared four computational methods for quantifying the effect of set-up error and uncertainty on delivered doses to targets and organs at risk in the intensity modulated radiation therapy treatment of head and neck cancer. These four methods were direct simulation, simple convolution, plus two modified convolution approaches. Discrepancies of up to 20% in the equivalent uniform dose (EUD) between direct simulation and simple convolution were estimated for the relatively superficial parotid gland at a systematic set-up error of 6 mm standard deviation and a random uncertainty of 2 mm standard deviation. Truncated convolution agreed with direct simulation to within 6% for all situations studied. However, of the four methods, only direct simulation can quantify the range of outcomes (EUD) associated with a finite number of courses and fractions. Our results are particularly relevant to the design of dose escalation studies in head and neck cancer.