Linear discriminant analysis is studied when the predictors are data of functional type(curves). Due to the infinite dimension of the predictor space, discriminant coefficients functions cannot be derived as in the classical setting.Partial least squares approach is then used in this context. Results of an application to kneading data are presented.