[Background] The imbalance of human energy homeostasis is characterized by underweight, overweight and obesity. Gut microbiota is related to the maintenance of human energy homeostasis, but the feature of gut microbiota in different body mass index (BMI) populations need to be further explored. [Objective] To explore the feature and the co-occurrence network of gut microbiota of four BMI groups based on the American gut project (AGP), and to provide new theoretical basis for intervention of obesity and underweight based on gut microbiota. [Methods] Screen the gut microbiota samples with BMI information from AGP dataset, and divide the screened samples into four categories according to the BMI classification standards set by the World Health Organization:underweight (BMI<18.5 kg/m²), normal weight (18.52), overweight (252), obesity (BMI>30 kg/m²). By calculating and comparing the alpha diversity and beta diversity of gut microbiota, the overall feature and differences of gut microbiota in four BMI groups were explored. Use Microbiome Multivariable Association with Linear Models (MaAsLin) to analyze the correlation between different BMI groups and gut microbiota, and add geographical, age, and gender factors as confounding factors to the model for correction. SparCC was used to calculate the correlation of gut microbiota among the four BMI groups, and the gut microbiota co-occurrence network of four groups was constructed. [Results] The decreasing trend of alpha diversity in the underweight, overweight and obese population compared to the normal weight population was observed after the Wilcoxon-rank-sum analysis. As for the beta diversity, the significant difference of the structure of gut microbiota between the four BMI groups was found. There was no significant difference in the relative contents of Firmicutes and Bacteroidetes. Through MaAsLin analysis, and adding geographical, age, and gender factors as confounding factors to the model for correction, a total of 49 bacteria that were significantly related to BMI types were obtained. The topological structure of the co-occurrence network of the four groups of BMI populations were different. The networks of underweight and normal-weight population were more complex, while the networks of overweight and obese populations were less complex. [Conclusion] There were differences in the diversity, overall feature and co-occurrence network of the gut microbiota among the four BMI groups.