Each crop variety has a genotype-specific ability to maintain performance over a wide range of environmental conditions. This ability is usually referred to as the sensitivity or adaptability of a variety. Such an ability is an important property, because farmers naturally want to use varieties which perform well in their own fields. Assessing sensitivity has, however, proved difficult, because of problems involved in defining and measuring the wide diversity of natural environments. These problems often lead to split statistical analyses of trial data or statistical models including explanatory variables with no biological interpretation. That causes ambiguity in statistical inference and prediction. The present study shows how the latest advances in statistical research can be applied to overcome some of these difficulties. A key point is to use the conditional expectation of the yield given the environment as a latent explanatory variable. In this way the predicted yields of different varieties can be estimated at any expected environmental yield level. Discussion is restricted to yield data but similar methods can be applied to other performance characters. The Finnish statutory variety trial data are used to illustrate the methods and the results.