The computer aided-system for the breast density pattern classification was built based on the researches of the objective quantification which converts breast density into glandular rate employing the phantom of the synthetic breast-equivalent resin material, and the subjective quantification of radiologists' visual assessment with the method of analysis of paired comparisons employing the Thurstone-Mosteller model. The system consists of the two processes. In the first process, pixels of a digitized mammogram are converted into glandular rates using the neural network to which the exposure conditions and the breast thickness were inputted. In the second process, the pattern classification and the glandular rate computation were performed taking visual assessment into consideration by the neural network to which feature values of the histogram of the glandular rate image converted into gray level were inputted. As a result of receiver operating characteristics (ROC) analysis estimating the pattern classification reliability to visual assessment in 93 samples, the area A(Z) under ROC curve was 0.95 or more values in each pattern. In the computed glandular rate to visual assessment, the maximum absolute error was 13% and the average absolute error was 3.4%.