The effectiveness of selection for 3-9 week gain was examined in a population of rats with a history of past selection for high 3-9 week gain. Lines were selected for high (U line) and low (D line) 3-9 week gain with two replicates of each line. Two randomly selected lines were also kept, one originating from the same base population as the two selected lines (R line) and the other originating from a population that had been randomly mated for the previous 27 generations (C line). Two replicates of each of these lines were kept. After seven generations of selection, a randomly selected line (relaxed line) was formed from each of the two upward- and each of the two downward- selected lines. Results have been presented for 13 generations of selection. The environmental trend for 3-9-week gain, as indicated by the randomly selected R and C lines, was consistently negative in all four lines. Realized heritabilities calculated by deviating the response to selection from the trend in the R or C lines resulted in non-significantly higher values in the D lines than the U lines. Six generations of relaxation of selection indicated no effect of natural selection in the U lines or the D lines. The relative magnitude of the drift, error and common environmental variances were estimated by the methods given by HILL (1971). The estimates of these parameters then led to calculation of the degree of bias in the sampling variances of the realized heritability estimates. As was predicted by HILL (1971), estimates of the variance of realized heritabilities obtained by using standard regression techniques were less than those obtained using HILL'S formulae. The results are discussed in relation to other similar studies with rats and mice.