The optimal unit hydrograph is shown to be the weighted sum of each runoff component which constitutes the crosscovariance. Statistical properties of hydrological data such as autocovariance, crosscovariance, spectrum, coherence and phase angle are investigated to apply the prediction method practically. It is shown that the autocovariance of daily rainfall may be roughly approximated by the Dirac delta function. The crosscovariance of rainfall and runoff analyzed in this paper is composed of two runoff components of short and long periods. The coherence of rainfall and runoff is nearly unity for light rainfall suggesting that the rainfall-runoff process may be approximated by a linear system. On the other hand, the coherence becomes worse as the rainfall intensity increases; there is a gap of coherence in the middle frequency range. Examples of runoff prediction by the theory are given.