For better preparing future epidemic/pandemic, important lessons can be learned from how different parts of China responded to the early COVID-19 epidemic. In this study, we comparatively analyzed the effectiveness and investigated the mechanistic insight of two highly representative cities of China in containing this epidemic by mathematical modeling. Epidemiological data of Wuhan and Wenzhou was collected from local health commission, media reports and scientific literature. We used a deterministic, compartmental SEIR model to simulate the epidemic. Specific control measures were integrated into the model, and the model was calibrated to the recorded number of hospitalized cases. In the epicenter Wuhan, the estimated number of unisolated or unidentified cases approached 5000 before the date of city closure. By implementing quarantine, a 40% reduction of within-population contact was achieved initially, and continuously increased up to 70%. The expansion of emergency units has finally reduced the mean duration from disease onset to hospital admission from 10 to 3.2 days. In contrast, Wenzhou is characterized as an emerging region with large number of primarily imported cases. Quick response effectively reduced the duration from onset to hospital admission from 20 to 6 days. This resulted in reduction of R values from initial 2.3 to 1.6, then to 1.1. A 40% reduction of contact through within-population quarantine further decreased R values until below 1 (0.5; 95% CI: 0.4–0.65). Quarantine contributes to 37% and reduction of duration from onset to hospital admission accounts for 63% to the effectiveness in Wenzhou. In Wuhan, these two strategies contribute to 54% and 46%, respectively. Thus, control measures combining reduction of duration from disease onset to hospital admission and within-population quarantine are effective for both epicenters and settings primarily with imported cases.