Until recently frequency-domain subspace algorithms were limited to identify deterministic models from input/output measurements. In this paper, a combined deterministic-stochastic frequency-domain subspace algorithm is presented to estimate models from input/output spectra, frequency response functions or power spectra for application as experimental and operational modal analysis. The relation with time-domain subspace identification is elaborated. It is shown by both simulations and real-life test examples that the presented method outperforms traditional frequency-domain subspace methods.