Leaf area index (LAI) is an important biophysical parameter for assessing of agroecosystems, which is widely used in various applications. The ground-based hyperspectral remote sensing technique is known to be inexpensive but effective for monitoring of the LAI of crop canopies. During the past twenty years period, hyperspectral technique has been adopted increasingly for plant LAI evaluation, which demands unique technique procedures compared with the conventional multispectral dataset, such as dimension reduction and denoising. Thus, identifying of the optimal bandwidths as well as effective wavelengths (sensitive wavelengths) is of great importance for improving the accuracy of crop LAI assessment based on the hyperspectral remote sensing data. As one of the most important oil crop in China, with a cultivated area of 7.5 million hectares and a production of about 14.4 million tons of seeds. Accurate and real-time assessment of spatial and temporal variations of crop LAI is particularly important. The objectives were to identify the optimal bandwidths and their effective wavelengths which were best suited for characterizing the winter oilseed rape biophysical variables. Five nitrogen field experiments involving different ecological sites, cultivars and planting patterns were carried out over three consecutive growing years (2013—2016) in Hubei, China. The in-site canopy hyperspectral reflectance dataset of winter oilseed rape were obtained over a wavelength region from 400nm to 1350nm (the visible and nearinfrared region), and quantitative correlations between LAI and their hyperspectra were analyzed. Moreover, a partial least square (PLS) regression model for LAI prediction was employed with different bandwidths (narrow and broad band spectral variables) canopy raw spectral reflectance (R) and its transformation technique: the first derivative reflectance (FDR). The prediction accuracy of the optimal bandwidths were determined by comparing coefficient of determination (R2), root mean square error (RMSE) and relative percent deviation (RPD) between the observed and predicted LAI values for both the calibration(cal) and validation(val) datasets. The results indicated that the values of LAI had a similar range in both the calibration dataset and the validation dataset and provided high variable coefficient values, indicating that the data partitioning was reasonable and could avoid unbiased evaluation. Compared with the R-PLS model for LAI estimation, the FDR-PLS model yielded higher retrieval accuracy for LAI prediction, and the optimal bandwidth was 20nm. The R2val, RMSEval and RPDval between the observations and predictions were 0.779, 0.414 and 2.004, respectively. The VIP scores of the FDR-PLS model with a full hyperspectral region (400~1350nm) were applied to select the effective wavelengths and decrease the high dimensionality of the canopy spectral reflectance data. Five wavelengths centered at 759nm, 847nm, 921nm, 1002nm and 1129nm were selected as sensitive wavelengths for monitoring the LAI status. The newly-developed FDR-PLS models for LAI prediction (R2val was 0.715, RMSEval was 0.486 and RPDval was 1.707) provided accurate estimations based on the field experiment validations using the effective wavelengths. The analytical thinking could provide an inventive thought thread of plant spectral wavelength selection for crop LAI prediction, and it also could provide a theoretical foundation for wavelength settings of broadband multispectral imaging spectrometer and monitoring potential applications of remote sensing data.