In model updating problems, one tries to find design variable values affecting the input to a numerical or analytical model, such that the model output is well correlated to some target response or experimental data. Because most proposed solution techniques are iterative or require large matrices to be inverted and multiplied, severe obstacles arise when large models, such as, for example, typical industrial finite element models, are at hand. Here we consider finite element model updating for structural dynamic applications and show how the problem with little effort may be projected onto a subspace spanned by, for instance, a few eigenvectors of the initial model. Because the subspace dimension is much smaller than the space spanned by the finite element basis, the subsequent correlation can be conducted speedily, although large changes from the initial design may require the subspace basis to be updated. The proposed method can conveniently be used for substructured models and, in addition, makes interactive design changes possible. A small model problem and an industrial application are used to illustrate the performance of the suggested approach.