Abstract
In Cassel and Lundquist (1990) the existence of sampling bias in estimating autocorrelation functions was discussed under a superpopulation model. One restriction of that model was that the time series model should not exhibit trend. In this paper we relax that restriction. The bias of a traditional estimator of the autocorrelation function is general of rather complicated nature. It depends on the sampling design, the time series model and the length of the sequence.