A linear-log model to predict yield loss due to metribuzin injury was established by Love et. al. in 1993. Two experiments were conducted in 2016 and 2017 to evaluate and improve this model for application in North Dakota (ND). Metribuzin was applied (1.12 a.i./ha) when potato plants were 20-30 cm tall at Inkster, ND. The model did not accurately predict yield loss in 2016 but performed better in 2017. Foliar injury was more correlated with yield reduction than relative plant height. Results also indicated that new models that used foliar injury at 21 days after treatment (DAT) data and at 7 DAT data, most accurately predicted total yield loss and marketable yield loss, respectively. The new model performed similar to the previous model, but unlike previous model it can predict yield loss very early in growing season (21 DAT).