A mathematical technique is described for modeling the power density spectrum of a random process from its periodogram. Advantages realized are the following. 1) The parameters are calculated in closed form not requiring any iterative computations. 2) Their variances decrease inversely as the time of observation increases. 3) The method provides a good rational approximation when the true spectrum

is basically of a nonrational form, such as when

is flat and band limited. The development utilizes a result on linearly dependent pencils of functions derived in part I.