If the test data is a short data sequence, it is not suitable to perfo rm conventional nonparametric estimation of frequency response function (FRF). Aimed at the problems, a new FRF estimation using the parameterized model was pr esented. A common denominator transfer function model was employed and the input and output spectrums were used as the primary data. The parameters were solved by the least square estimator. The technology of singular entropy was adopted to determine the order of system. A simulation case of a 2storey building demons trates that the algorithm is effective.