This study uses the choice experiment method with 570 grain family farms located in the Huang-huai-hai Plain and determine various support policy attributes and the attribute levels for the two dimensions of policy measures and policy communication channels. Ordering effects are eliminated by warming up subjects in advance and using information disclosure. This paper uses the inferred attribute non-attendance method to process attributes ignored by the grain family farms and analyzes grain family farms’ preferences for different support policies with a mixed logit model and then uses a latent class model to analyze how the characteristics of grain family farms relate to different preference types. We find that grain family farms have a strong preference for agricultural subsidies, credit support, and technical support (the mean coefficient is greater than 0.8). Moreover, the preferences of grain family farms over the policy communication channel (the mean coefficient is greater than 0.5) cannot be ignored. Faced with the same policy attribute combination, grain family farms with high education levels, reasonable scales of operation, and good understanding of support policies are more likely to improve their profit margins. There are four preference types of grain family farms: finance preference (43.2%), knowledge and technology preference (28.5%), land transfer preference (15.4%), and policy information preference (12.9%).