SUMMARYIn this paper, we deal with a numerical realization, which is a numerical analysis methodology to reproduce real ows by integrating numerical simulation and measurement. It is di cult to measure or calculate ÿeld information of real three-dimensional unsteady ows due to the lack of an experimental ÿeld measurement method, as well as of a way to specify the exact boundary or initial conditions in computation. Based on the observer theory, numerical realization is achieved by a combination of numerical simulation, experimental measurement, and a feedback loop to the simulation from the output signals of both methods. The present paper focuses on the problem of how an inappropriate model or insu cient grid resolution in uences the performance of the numerical realization in comparison with ordinary simulation. For a fundamental ow with the Karman vortex street behind a square cylinder, two-dimensional analysis is performed by means of numerical realization and ordinary simulation with three grid resolutions. Comparison of the results with those of the experiment proved that the feedback of the experimental measurement signiÿcantly reduces the error due to insu cient grid resolution and e ectively reduces the error due to inappropriate model assuming two-dimensionality.