A general method is developed for forming fractionally replicated designs for mixed series. Designs are given for half-replicates of the 24 x 3, 24 x 4, and 23 x 3 x 5 factorials. Least squares estimates of the effects are discussed. The analysis of variance for the 24 x 3 and 24 x 4 half-replications and the variances of the estimates of the observations not taken under the design are compared. A numerical example is given in which the results of analyzing data from a full 24 x 3 experiment is compared to the analysis of a pretended half-replicate.