The need for sustainable agriculture requires the adoption of low input, long-term and cost-effective strategies to overcome the adverse impact of disease and nutritional deficiencies on citrus groves. In this context, early detection of diseased trees has become an important topic in the citrus industry. Multiple factors make field assessment of disease conditions a challenging task: the non-specific nature of many symptoms, the possibility of having localized affections in only certain areas of the tree or the correlation with other factors such as tree age. In this paper we investigate hyperspectral sensing as an effective approach to detect the Huanglongbing disease in citrus trees. We analyze the visible and near infrared spectral responses from the leaves to discriminate infected trees from healthy samples. The accuracy of the diagnosis is improved by means of feature selection techniques that prevent overfitting problems due to the high dimensionality and collinearities in the data. We provide experimental results illustrating the performance of the proposed techniques using data collected in the field.