As the world's largest terrestrial ecosystem, forest is very important to human living and environment sustainable development. Therefore, grasping the status and changes of forest resources are of significance. But classification of sub-category information of forest vegetation has always been difficult for remote sensing, because of the impact of complex terrain, irregular distributed vegetation, and the similar spectral information of different forest types. In recent years, classification combining spectral characteristics and multivariable remote sensing data has particularly become study focus. In this study, Eastern Jilin was chosen as the study area, where approximately 80% of the land is covered with forest vegetation, and the sub-category of forest vegetation contained broadleaved deciduous forest, deciduous coniferous forest, evergreen coniferous forest, mixed broadleaf-conifer forest, and deciduous shrub. The classification was operated based on object-oriented method using HJ-1 CCD data and MODIS-NDVI data. A hierarchical segmentation method was proposed in this paper. Different segmentation parameters could be set according to different land cover types. Firstly, non-forest land cover types were classified. Then the sub-category of forest vegetation was classified based on the characteristics of the spectral features generated by HJ-1 CCD data, and phenological features generated by MODIS-NDVI time series data. Among these sub-forest vegetations, the broadleaved deciduous forest and deciduous shrub, the evergreen coniferous forest and deciduous coniferous forest are similar in spectral features, but obvious different in phenological features. In this study, the spectral features used to classify sub-forest vegetation are Mean Layer 2 (mean value of HJ-1 CCD band 4), Mean NDVI; the phenological features including Mean Layer 5 (mean value of MODIS-NDVI 81d), Mean Layer 8 (mean value of MODIS-NDVI 129d). There are 707 ground truth points used to assess the classification accuracy, including 622 forest points and 85 non-forest points. The overall accuracy is 91.5% and Kappa confidence is 0.88, the broadleaved deciduous forest got the highest accuracy, the producer's accuracy is 97.1% and the user's accuracy is 92.1%, other sub-categories of forest vegetation all got accuracy approximately 90%. In order to compare the classification results with and without MODIS-NDVI time series data, we chose a small area to operate the object-oriented classification without MODIS-NDVI time series data. The comparison indicated that without MODIS-NDVI time series data the classification image appears very disordered. Among the forest sub-categories, deciduous coniferous forest and evergreen coniferous forest, broadleaved deciduous forest and deciduous shrub are remarkablely mixed. The classification accuracy is also quite low, the overall accuracy is 61.5% and the Kappa confidence is 0.53. The comparison ensured that the joined of MODIS-NDVI time series data significantly improved the forest sub-categories' classification result. The classification method operated in this study (based on object-oriented method combining HJ-1 CCD data and MODIS-NDVI data) could also be used in classifying vegetation in other regions, but the parameters in this study is regional adoptive.