In this paper, the development and integration of sensor technology for herring roe grading is described. In this application, by and large, grading is presently done by hand. Shape, firmness/texture, color and weight all play key roles in determining the grade of a roe skein. Furthermore, a number of these variables are quite subjective both in the way they are determined by human graders, and in the way they affect the overall grade of a skein. Thus, in the present system an "intelligent" decision-making system, with multiple levels of abstraction, is used to determine the final grade of the product. The system uses a model base and also a knowledge base developed through interactive acquisition of knowledge from grading experts and off-line experiments. These are implemented as a set of fuzzy-logic rules and model matching procedures. A prototype grading machine has been developed and tested at an industrial plant. Topical experimental results are presented and studied.