The purpose of this study is to combine "item response theory of non-parameter", "fuzzy theory" and "latent class model" to develop a newintegrated model called "fuzzy latent knowledge space". The integrated model is suitable for analyzing psychometrical data. In addition to having the traditional features and meanings of knowledge space, the integrated model also improves some defects of Yamamoto(1987) and Tatsuoka(1995), and enhances the method of Yamashita, Katsumata, & Tsuda(1994).Therefore, the features of the integrated model are as follows: (1)showingthe membership of task-takers with respect to each latent class; (2)showing the developmental relationship between each latent class by approximate ternary graph; (3)showing the similarity structure of items in each latent class by partition tree graph; (4)showing the ideal response pattern, ability,and distance of master of each latent class; (5)the items could be dichotomous,polytomous, or mixed scoring; and (6)there is no restriction of "local independence" between items. There are two major parts in this study. They are real data analysis and data simulation. As to the real data analysis, the researcher uses two datasets of "indicators of mathematics learning progress" and "concepts ofpartitive word problems". We can realize the features of the integrated modelaccording to the process of analyzing data. As to the data simulation, under the manipulated two factors-- "distance of mean" and "proportion of population",we can understand the characteristics of fuzzy partition. According to the real data analysis and the data simulation, the inuegrated model could be developed for tool of computerized cognitiondiagnosis or neural networks cognition diagnosis, which is very useful for psychom etrics. Based upon the findings of this study, recommendations for further research are suggested.