We propose a new approach, called Extreme-PLS, for dimension reduction in conditional extreme values settings. The objective is to find linear combinations of covariates that best explain the extreme values of the response variable in a non-linear inverse regression model. The asymptotic normality of the Extreme-PLS estimator is established in the single-index framework and under mild assumptions. The performance of the method is assessed on simulated data. A statistical analysis of French farm income data, considering extreme cereal yields, is provided as an illustration.