(eng): We are concerned with Bayesian approach to data assimilation. Many common assimilation techniques, such as Kalman filtering and particle filtering, are special cases of the approximative solution of the Bayesian equations. Both of these techniques use a single approximation for the posterior probability density function (pdf) of the state vector. In this paper we advocate the use of marginalized particle filter (MPF, also known as Rao-Blackwelized particle filter), which splits the state vector in two disjoint parts and approximates their densities differently. Simulation study was performed as a twin experiment when measurements are sampled from a twin model and perturbed. The resulting assimilation scheme proved to be robust with respect to initial conditions and the inherent adaptivity of the inflation factor accelerated convergence to the simulated values.