The cyclic autocorrelation is often used to describe nonstationary ran
dom processes. In this paper we investigate the conditions under which
the cyclic autocorrelation can be estimated consistently in mean squa
re for discrete time Gaussian processes. We extend and generalize resu
lts of Hurd [17] and refine results of Boyles and Gardner [1]. We deri
ve necessary and sufficient conditions for consistency in mean square
of an estimator, which are in the form of a single sum of autocorrelat
ion coefficients, in the form of a double sum of autocorrelation coeff
icients, in the bifrequency domain and in terms of the average spectru
m. We also discuss the rate of convergence for this estimator.