A method was proposed for the identification of citrus huanglongbing by near infrared(NIR) spectroscopy based on least angle regression combined with kernel extreme learning machine(LAR-KELM(RBF)) as the traditional detection method for the disease has some defects such as low accuracy and poor stability.Firstly,the acquired spectral data were preprocessed by wavelet transform,then the least angle regression(LAR) algorithm was used to select the spectral wavelength,and finally,with the help of KELM(RBF),the filtered spectral data were managed to classify.The NIR spectral data of orange leaves were taken to verify the performance of LAR-KELM(RBF) algorithm in the experiment.The classification accuracy of the algorithm could reach up to 99.91%,and standard deviation(STD) was 0.11.The experimental results of different training sets showed that LAR-KELM(RBF) model was more accurate and stable than extreme learning machine(ELM),summation wavelet extreme learning machine(SWELM),back propagation(BP(two layers)),KELM(RBF) and support vector machine(SVM) model,which could be widely used in the detection and differentiation of citrus huanglongbing.