In this work, NIR and MIR spectroscopy was investigated and compared for predicting individual sugars (sucrose, glucose and fructose). Spectra of 56 samples of lyophilized passion fruit were collected using an integrating sphere in NIR range and attenuated total reflectance accessory in MIR range. Spectral and reference data were analyzed by principal component analysis. Partial least square regression (PLSR) was used to establish calibration models. MIR technique was better than the NIR technique for glucose (R 2 v = 0.942), fructose (R 2 v = 0.855) and sucrose (R 2 v = 0.818).