A three-dimensional circulation model of the Chesapeake Bay is used to validate a simple data assimilation scheme, using high-resolution salinity data acquired from a ship-towed undulating vehicle (a Scanfish). The simulation period spans the entire year of 1995 during which the high-resolution Scanfish data were available in July and October, lasting a few days each. Since Scanfish data were irregularly distributed in time and space, only salinity fields are nudged in the model for simplicity. Model improvements through data assimilation are evaluated from a pair of experiments: one with data assimilation and one without. Data from scattered Chesapeake Bay Program monitoring stations and a few stations maintained by the National Ocean Service inside the bay are used independently to check the model performance. In general, the simple assimilation scheme leads to visibly improved density structures in the upper and middle reaches of the bay. The improvement in the lower bay is equally pronounced after data assimilation but diminishes in a shorter timescale because of faster flushing from the adjacent coastal ocean. Moderate to weak nudging normally enhances the gravitational circulation. Strong nudging may produce transient overshooting, during which the gravitational circulation is renewed vigorously.