Previous research using NTEP data has shown the additive main effect and multiplicative interaction (AMMI) model to be more accurate than the cell means model (means averaged over replicates) in predicting future performance. This study was developed to validate in the field the accuracy of AMMI models (AMMI adjusted means) and standard NTEP models (AMMI-F, cell means) for predicting turfgrass quality (TQ). AMMI analysis of the 1995 NTEP Kentucky bluegrass test (4-year averages) was used to develop winning rosters for planting at various test locations. Five winners as suggested by AMMI-5 adjusted means (AMMI winners) and five winning genotypes based on means averaged over replicates (AMMI-F, NTEP winners) were included in the test roster. Winning rosters (ten entries) were planted at six test locations in the fall of 2003 and TQ data (1 to 9 scale, 9=highest) was submitted by cooperators in 2004 and 2005 and then averaged. AMMI-5 adjusted means and actual TQ (2004-2005 averages) were significantly and positively correlated at two of the six test locations including Ames, IA (r=0.67, p = 0.05) and Adelphia, NJ (r=0.89, p = 0.001). AMMI-F cell means were negatively correlated with actual TQ at two test locations including Blacksburg, VA (r= -0.66, p  0.05) and Adelphia, NJ (r= -0.60, p = 0.05). AMMI-F predictions at these locations were costly since predicted winners were losing genotypes. Compared to AMMI-5, AMMI-F predictions over-estimate statistical gains in actual TQ in planting AMMI-F winners over winning entries suggested by the AMMI-5 model. The AMMI-5 model was twice as likely to recommend winners (rank 1 and 2) than AMMI-F while the AMMI-F model was twice as likely as AMMI-5 to recommend losing entries (rank 9 and 10). The AMMI-5 model predictions of actual TQ were more accurate than AMMI-F (cell means).