We propose a test for separability of the correlation structure of a multivariate time series. We construct test statistics based on a spectral density matrix estimated in a nonparametric way and derive their asymptotic properties. We use simulation to check the performance in finite samples