The performance of reflectance spectral measurement using a spectroscopic method could be affected by scanning distances. Therefore, this study was undertaken to investigate the effect of scanning distances on the performance of a spectroscopic method to predict oil content of oil palm fruitlets. A total of 216 fruitlet samples from six bunches at different maturity stages were used. Spectral data was collected using a high-resolution fibre optic spectrometer with wavelengths ranged between 500 and 900 nm. The samples were scanned at five different scanning distances namely 0, 1, 2, 3, 4 and 5 cm. Soxhlet extraction analysis was performed to determine the oil content of the samples. Partial least square (PLS) regression method was used to develop calibration and prediction models to correlate the spectral data with the respective oil content of the samples. For prediction models, the coefficient of determination (R2) for 0, 1, 2, 3, 4 and 5 cm scanning distances were 0.95, 0.93, 0.86, 0.83, 0.80 and 0.80, respectively. Artificial neural network (ANN) was used to classify fruitlets samples into the maturity stages and yielded good classification performance with an average of 94%. These results indicate that the oil content of fruitlet samples can be predicted by using the spectroscopic method. However, the scanning distances could affect the prediction accuracy of the models. This study has also demonstrated that the spectroscopic method coupled with ANN algorithm could be applied to classify the maturity stages of oil palm fruitlets.