Abstract: Plant height is an important indicator of the rice growth status. Current remote sensing retrieval methods for plant height are limited by data cost and update frequency and cannot meet the requirements of large-scale and multi-frequency operational monitoring of crop plant height; therefore, there is still a need to fully explore the potential of high spatial and temporal resolution optical satellite data. Choosing rice in the northeastern region of China as the research subject, measured plant height data from 2022 and 2023 were used, and an optical image-based feature set was constructed using Sentinel-2 data. Based on this, statistical analysis was used to explore the sensitivity of each feature to plant height. Multiple regression models with different feature combinations were then constructed using the random forest and extreme gradient boosting algorithms. Simultaneously, feature recursive elimination and importance-ranking methods were employed to analyze the importance of single features and different categories of features. Finally, the overall potential of Sentinel-2 images was evaluated. Finally, the random forest model with all optimized features achieved the highest estimation accuracy, with R²=0.857 and RMSE=8.395 cm. Comparative analysis revealed that among the individual features, the vegetation water index and blue band were highly important. Red edge features, red bands, green bands, and texture means were also selected into the optimal feature set, demonstrating certain plant height monitoring capabilities. Among the different categories of optical features, the vegetation index features were more important than the reflectance features. When texture features were applied alone, it was difficult to achieve a high estimation accuracy, but they could serve as supplementary information when combined with the other two categories of features to improve the estimation accuracy. Compared with previous studies, the accuracy of plant height estimation using Sentinel-2 data is comparable; however, it enables rapid, multi-period, and large-scale estimation of plant height, showing good application potential.