Numerical models used to forecast the atmospheric dispersion of radionuclides following nuclear accidents are subject to substantial uncertainties. Input data, such as meteorological forecasts or source term estimations, as well as poorly known model parameters contribute for a large part to this uncertainty. A sensitivity analysis with the method of Morris was carried out in the case of the Fukushima disaster as a first step towards the uncertainty analysis of the Polyphemus/Polair3D model. The main difficulties stemmed from the high dimension of the model's input and output. Simple perturbations whose magnitudes were devised from a thorough literature review were applied to 19 uncertain inputs. Several outputs related to atmospheric activity and ground deposition were aggregated, revealing different inputs rankings. Other inputs based on gamma dose rates measurements were used to question the possibility of calibrating the inputs uncertainties. Some inputs, such as the cloud layer thickness, were found to have little influence on most considered outputs and could therefore be safely discarded from further studies. On the contrary, wind perturbations and emission factors for iodine and caesium are predominant. The performance indicators derived from dose rates observations displayed strong sensitivities. This emphasises the share of the overall uncertainty due to input uncertainties and asserts the relevance of the simple perturbation scheme that was employed in this work.