The z-test for treatment effects in randomized blocks and Latin square exps. is discussed. The distribution of z generated by randomization on the "null" hypothesis is compared with the z-distribution of normal theory. The comparison is made through the medium of a monotonic function U of z. The mean U on randomization agrees with the normal theory value, but the variance differs. The results are applied to uniformity trials. The discrepancy between the randomization and normal theories was not serious for the randomized block trials considered; for the 3 Latin square trials, the randomization variance of U was appreciably smaller than the normal theory value. Application to a wider range of trials is suggested as useful.