At early stages in cultivar development, genotypes are usually evaluated in a single replication because: 1) seed supplies are too limited, and 2) the number of genotypes to evaluate is too large to permit replicated evaluation. One method that has been proposed for conducting unreplicated tests is the Modified Augmented Design (MAD) which utilizes repeated check genotypes to measure and adjust for field variability and to estimate error variance. Use of MAD has not been reported in cotton (Gossypium hirsutum L.), and it is uncertain how effective the approach is in practice in removing the effects of field non-uniformity. Two types of experiments were conducted to evaluate the utility of MAD. Experiment 1 consisted of a MAD superimposed on a randomized complete block design (RCBD) at three Mississippi locations. This allowed the comparison, on a replicated basis, of unadjusted yields and yields adjusted by MAD. In uniform sites, MAD was not effective in reducing error or improving the ratio of genetic and error variances; in a non-uniform site, MAD was effective in this regard. Experiment 2 consisted of four large (360 to 540 entries) tests using MAD. Based on standard deviation of appropriate check plots, error variance was reduced in all four of these tests. The reduction in error variance was generally sufficient to justify the commitment of 10-20% of plots to checks.