Abstract: To optimize the distribution of light source in the aspect of non-invasive analysis of soluble solids content (SSC) by visual/near-infrared (Vis/NIR) transmittance, four arrangements of light source distribution were designed to diminish the influence of the random acquisition-spot on ‘Red Fuji' apple by USB2000+portable fiber spectroscopy. After getting rid of the outliers and several pre-treatments on the transmitted spectra, partial least square (PLS) models were built between the spectra and the SSC, which was determined by regular method. Comparison of models ' root meant standard error prediction (RMSEP) and correlation coefficient (rp) , it was found that the type of that low angle of light illumination combined with multi-light sources was the best one to show the internal qualities of the fruit, with the corresponding prediction model for SSC as rp=0.804, as well as RMSEP=0.635. This might be a good reference to light-source distribution for online or portable device non-destructive determination used for industrial applications.