It is important for healthy aging to understand resilience in depth. This study aims to examine the dimensional structure underlying the Connor–Davidson Resilience Scale (CD-RISC) among Chinese older adults. Exploratory Graph Analysis (EGA) was used to evaluate the dimensional structure of CD-RISC in two large samples: training sample (n = 11,493) and cross-validation sample (n = 7662). Then, Confirmatory Factor Analysis (CFA) was used to compare the fit of the theoretical dimensions with the EGA dimensions. Finially, Generalized Linear Model was used to examine the association between resilience scores and self-rated health (SRH) after controlling other covariates in order to evaluate the predictive value of the EGA dimensions. The EGA indicated two demensions(named foresight and self-adjustment) of the 25-item CD-RISC. The CFA comparison found that the two-demension structure of CD-RISC fit significantly better than the theoretical three-demension structure. After controlling for sociodemographic characteristics, generalized linear model showed that the EGA dimensions has better protective value with SRH. Compared with older adults with lowest quartile of foresight, those with second (odds ratio, OR = 0.68, 95% CI = 0.62 ~ 0.75), third (OR = 0.50, 95% CI = 0.45 ~ 0.56) and fourth quartile (OR = 0.42, 95% CI = 0.37 ~ 0.48) of foresight had lower odds ratio of poor SRH. Similarly, older adults with the second (OR = 1.11, 95% CI = 1.01 ~ 1.23) and fourth (OR = 0.79, 95% CI = 0.69 ~ 0.90) quartile of self-adjustment also had lower OR of poor SRH than those with lowest quartile of self-adjustment. These findings show that EGA outperforms the traditional methods, which may be helpful to understand resilience deeply. CD-RISC should be interpreted into two aspects among community-dwelling older adults in China, highlighting the significance of the practical value and cultural context of resilience.