Abstract: Preference handling is one of the important research contents in the field of artificial intelligence. CP-nets (conditional preference networks) as a popular language capable of representing ordinal preference relations in a compact and structured manner, has received increasing attentions in recent years. A CP-net is an annotated directed graph that encodes the preference relationships among variables of interest. The conditional preference networks is a simple and intuitive graphical tool for representing conditional ceteris paribus preference statements over the values of a set of variables. But there are very few works that address the problem of structure learning of CP-nets. In this paper, the structure learning of CP-nets is studied, in particular, learning algorithm based on the G² test to learn the structure of CP-nets are discussed detailedly. Furthermore, it is proved that the time complexity of this algorithm is O (n·2n) . As a log likehood ratio testing approach, G² testing is especially suitable for testing the causal relationship between variables. Conditional preference independence is the core concept of CP-nets, so using G² test can learn the CP-nets structure effectively. By constructing G² test statistics, and performing null hypothesis test from the given pairwise comparisons samples, we can get the parents of each node, therefore obtain the structure of CP-nets. Finally, the effective of proposed algorithm is verified by random generated simulation data. Compared with other CP-nets learning approaches, the approach proposed in the paper has the characteristic of passive, offline and essence of statistical learning.