Infrared spectroscopy (IR) is a well-established technique for determining component levels in foods. New portable technology offers versatility and ruggedness by reducing the size and weight of the spectroscopy systems without compromising performance and reproducibility. The aim of this work was to evaluate the performance of regression models developed by dispersive NIR, FT-NIR and MIR using laboratory benchtops and portable instruments for determination of the main components in ground endosperm of common beans (protein, starch, and amylose). In order to increase the speed of data collection and prevent cross-contamination between samples, NIR spectra of samples in a polystyrene bag were taken. Endosperms of 54 different genotypes cultivated in different locations in Catalonia and Asturias (Spain) were analyzed by standard chemistry (reference) methods and their spectra were collected using benchtop and portable systems on dispersive NIR, FT-NIR and MIR. Partial least squares regression (PLSR) was used to correlate the spectra matrix with the reference value and the regression model was validated by full-cross-validation. Ratio of performance to deviation (RPD) value was used to test the regression model performance, with the FT-NIR benchtop system giving the best RPD values of 3.7, 3.2 and 2.8 for protein, starch and amylose levels, respectively. Handheld NIR and MIR portable systems showed good predictability with RPD values of 3.5, 2.9 and 2.6 for protein, starch and amylose levels, respectively. NIR spectral collection using polystyrene bags had detrimental impact in the PLSR models in terms of reduction of the models goodness of fit and RPD values. Portable infrared units may provide the plant breeder and germplasm banks with an efficient method for simple, rapid and high throughput quantification of quality parameters in common beans. © 2013 Elsevier Ltd.