Kalman filters are widely used for data assimilation into ocean models. The aim of this study is to discuss the relevance of these filters with high resolution ocean models. This was investigated through the comparison of two advanced Kalman filters: the singular evolutive extended Kalman (SEEK) filter and its ensemble-based variant, called SEIK filter. The two filters were implemented with the Princeton Ocean model (POM) considering a low spatial resolution configuration (Mediterranean sea model) and a very high one (Pagasitikos Gulf coastal model). It is shown that the two filters perform reasonably well when applied with the low resolution model. However, when the high resolution model is considered, the behavior of the SEEK filter seriously degrades because of strong model nonlinearities while the SEIK filter remains remarkably more stable. Based on the assumption of prior Gaussian distributions, the linear analysis step of the latter can still be improved though.