Abstract: There is a multi-level nested structure of “nation-regions (provinces) -industries-enterprises” in China’s economy, which has not been studied in previous decompositions of productivity. Based on the dynamic Olley-Pakes productivity decomposition model, by introducing the factors flow at regional and/or industrial levels, the effect of factor flow was decomposed into the selection effects of different levels, which can be applied to other related models. The results of the empirical analysis of China’s manufacturing in 1998—2007 are as follows: (1) The annual growth rate of value-added based total factor productivity (TFP) is 11.57%, which has contributed 58.43% to the manufacturing growth and has obvious heterogeneity among industries and regions. The relative contribution of enterprises’ learning effect is the largest (62.10%), followed by the factor flow effect (36.78%), but enterprises’ turnover effect is very small. (2) The fewer the hierarchy of factor flow in the model, the more enterprises’ learning effect (factor flow effect) is overestimated (underestimated). (3) When excluding outliers of TFP across all observations, the rapid growth and industrial heterogeneity of TFP will be ignored, which will underestimate the TFP growth of China’s manufacturing.