Early recognition of pathogen infection is of great relevance in precision plant protection. Disease detection before the occurrence of visual symptoms is of particular interest. By use of a laserfluoroscope, UV-light induced fluorescence data were collected from healthy and with leaf rust infected wheat leaves of the susceptible cv. Ritmo 2-4 days after inoculation under controlled conditions. In order to evaluate disease impact on spectral characteristics 215 wavelengths in the range of 370-800 nm were recorded. The medians of the signatures suggest that inoculated leaves may be separated from healthy ones based on fluorescence measurements. Noise, high-frequency oscillations and individual reactions of leaves indicate that separability is difficult to achieve. Above, we observe a misbalance between the high number of observed wavelengths and the low number of training examples which induces a high risk of overfitting. For an early identification of plant diseases a small number of robust features is needed which comprises most of the information relevant for the given classification task. To this end several approaches for dimension reduction and subsequent classification where evaluated. The fluorescence curve was approximated by splines. The resulting coefficients were used as features specifying global, quite robust curve characteristics. However, piecewise fitting by polynomials of fourth order led to superior results. Subsequently, different classification algorithms such as Decision Trees, Naive Bayes, Artificial Neural Networks, Logistic Regression and Support Vector Machines (SVMs) were compared. SVMs achieved the best results. Cross validation demonstrated that the achieved classification accuracy reached 93%. This result could be attained on the second day after inoculation, before any visible symptoms appeared. The described method is of general interest for the early, presymptomatic identification of plant diseases based on hyperspectral fluorescence signatures.