Carambola is a tropical fruit with rising value in developed countries due to its nutritional value and exotic aspect. It is important to assess carambola quality in different maturity stages to estimate a “fair price” and to assign fruit for specific applications and markets. This work reported the use of a portable NIR spectrometer in the range of 900 to 1700 nm as a non-destructive, chemical-free technique for determination of carambola physicochemical properties, according to maturity stage. Colour, total soluble solids, ascorbic acid, moisture, pH and titratable acidity analysis were performed for 177 fruit from two clones and four maturity stages (MS1, MS2, MS3 and MS4). PLS-DA and PLSR models were built to classify carambola according to maturity stage and to estimate its physicochemical properties, respectively. Several pre-processing were tested and among them the new algorithm introduced in the (SNV) pre-processing, the variable sorting for normalization (VSN), allows the improvement of the signal shape and the model interpretation. Genetic algorithm (GA) and interval partial least square (iPLS) were tested for improving model performance. The PLS-DA model based on important variables selected by iPLS achieved the best performance with 84.2% accuracy to classify carambola according to maturity stage. Variable selection (iPLS and GA-PLS) allowed an improvement in the performance of the PLSR models, with pH and moisture content achieving (R2P of 0.78 and 0.74), (RMSEP of 0.2 and 0.87), (RPD of 2.01 and 2.23) and (RER of 8.02 and 10.38), respectively, which is acceptable for screening. Portable NIR spectrometer, which can be considered low-cost when compared to benchtop spectrometers, in tandem with chemometrics can be a promising tool to assess the composition and to classify carambola according to maturity stage.