The non-parametric discrete-time estimation of the covariance function R(t) of stationary continuous-time processes is considered. The characteristics of the sampling instants necessary for the consistent estimation of R(t) are explored. A class of covariance estimates is introduced and its asymptotic statistics are derived.