This paper designed an intelligent on-line hierarchical device to grade apple according to its sugar content on the basis of the principle of spectroscopy, and mainly studied the stray light problem of the on-line equipment with receiver at the top of the line. By analyzing the diagram of light path, we analyzed the influence of three different laying ways of Fuji apples on the stray light and put forward a shading treatment to reduce the noise of stray light. The models for on-line predicting soluble solids content (SSC) of Fuji apples were built respectively by using the partial least squares (PLS) regression. The results proved that, for online equipment with receiver at the top of the line, the laying way of apple and the shading treatment were very important. Among these laying ways, “peduncle up and shading treatment” was the best. The parameters of the model are pretty good, with RMSEC of 0.60, RC of 0.94, RMSEP of 0.67, and RP of 0.87. It’s able to achieve the demands of grading apples on-line in practical production.