The aim of this study was to compare a laboratory-based and a pocket-sized near-infrared (NIR) spectrophotometer (SCiO) to predict mango fruit firmness. Three batches of mango fruit were measured to explore batch specific and global models. To gain insight to useful wavelengths important for predicting mango firmness variables, the bootstrapping soft shrinkage (BOSS) variable selection routine was used. Modelling was performed with partial least-squares regression (PLSR). The reference firmness measurements were performed with AWETA acoustic firmness analyser. The SCiO and the laboratory-based instrument showed similar performances predicting the reference firmness in terms of prediction coefficient of determination (R2). However, the root mean squared error of prediction (RMSEP) was slightly lower for the laboratory-based instrument compared to the SCiO, likely because a broader spectral region was used. The performance of the batch specific models was improved by up to 8% in R2 with a 13% reduction in RMSEP when BOSS was applied. For both laboratory based and SCiO, the global models based on combined data from the three batches, showed good performance (R2 0.74–0.93, RMSEP 4.8 – 8.2 Hz2g2/3 depending on the batch) to predict the firmness. Due to comparable performance of the SCiO compared to the laboratory-based spectrophotometer, the pocket-sized SCiO NIR sensor has the potential to become a low-cost, easy to use non-destructive tool to measure firmness in mango fruit.