We propose a new method to improve the utility of three-dimensional ocean circulation models. The method uses climatological temperature and salinity data to adjust the momentum balance of the model, while leaving the tracer equations fully prognostic and unconstrained. The adjustment is accomplished by replacing density in the hydrostatic equation by a linear combination of model-computed and climatological density. The procedure is equivalent to adding a forcing term to the horizontal momentum equation through a modification of the model's horizontal pressure gradient term. The forcing term modifies the model-computed velocity field, which, in turn, affects the model-computed temperature and salinity fields through the advection term (there is no adjustment of the tracer equations carried by the model). Assuming the linear combination coefficient to be invariant in time and space, we suggest a statistical approach to estimating its optimal value. We apply this semiprognostic method to the northwest Atlantic. A primitive equation circulation model is initialized with January climatological temperature and salinity and is forced by monthly mean Comprehensive Ocean-Atmosphere Data Set surface wind stress and heat flux, by restoration of the surface salinity to monthly mean climatology, and by flows through the open boundaries. Both the model-computed tracer and velocity fields produced using the semiprognostic method show significant improvement over those produced by a purely prognostic calculation; drift of the tracer and velocity fields away from climatology is greatly reduced. Further, convective mixing is explicitly represented, thus improving the utility of results over those obtained from pure diagnostic calculations. The velocity fields obtained with the new approach are somewhat more realistic than those obtained from pure diagnostic calculations. The method reproduces many well-known circulation features in the region, including the Labrador Current, the Gulf Stream, and the North Atlantic Current. More significantly, the method reproduces reasonably well the seasonal evolution of temperature and salinity in the region despite the fact that the model's tracer fields are not constrained directly by the new method. This result suggests that the semiprognostic approach will be useful for examining the evolution of tracers that are not easily determined by observations. ¿ 2001 American Geophysical Union