When GNSS height determination improves in the future, users will ask for increasingly better geoid models. It is not unlikely that a standard error of 5 mm will more or less be required in a couple of years. The main purpose of this paper is to investigate the gravity data requirements to compute a Swedish gravimetric quasigeoid model to that order. The propagation of errors in the terrestrial gravity observations and the Earth Gravitational Model (EGM) are studied using both variance-covariance analysis in the spectral domain and least squares collocation. These errors are also checked by computing a new gravimetric quasigeoid model and comparing it with GNSS/levelling height anomalies. It is concluded that it will be possible to compute a 5 mm model over Sweden in the case that the gravity data set is updated to fulfil the following requirements: the resolution should be at least 5 km and there should be no data gaps nearby. Finally, the standard errors of the uncorrelated and correlated gravity anomaly noises should be below 0.5 and 0.1 mGal, respectively.