Approximate modeling of discrete-time linear time-varying systems is studied based on a representation of linear systems in the Gabor time-frequency space. The time-varying system is assumed to be given in input-output or kernel representation. This representation has recently received attention because of its applicability in frozen-time analysis and design of optimal control for time-varying systems but requires a large number of coefficients. Motivated by its application to signal analysis, the Gabor transform is considered as a tool for the representation and approximation of linear time-variant systems. In order to show the main results, the class of systems considered is restricted to the one usually considered in the frozen-time approach. An example is included to illustrate the potential application of the technique.