The use of high-gains in the filtering problem for non-linear stochastic systems is studied. It is shown that a high-gain filter has the same structure as a Luenberger observer. The use of high-gains is studied in the cancellation of non-linearities in order to simplify the filter design for non-linear systems. Singular perturbation techniques are used to show how the error dynamics reaches stable equilibrium in a very fast transient time, ensuring that the slow dynamics of the filter are just those of the non-linear stochastic system.