By choosing agricultural input and output panel data from 31 provinces and municipalities from 1999 to 2015，the Global Malmquist Luenberger Productivity Index (GML) method was used to measure the growth and decomposition of agricultural green total factor productivity (TFP) in China and to test its convergence. The results show: (1) China's agricultural green TFP growth is due to technological progress, scale efficiency reduces the increase in green TFP, pure technical efficiency has no obvious effect; (2) Failure to consider carbon emissions from agriculture will overestimate the level of TFP, leading to deviation in the evaluation; (3) Considering the carbon emissions from agriculture, the TFP levels show decreasing trend in the east, middle, west and northeast; (4) There are no obvious σ convergence characteristics in the whole China and regions, absolute β convergence exists at the national level and in the northeast, there are significant conditional β convergences across the country and in the east, middle, west and northeast regions, it shows that all provinces have developed toward their own stable state, while the "catch-up effect" between provinces is not obvious. Based on this, it is believed that agriculture should take the path of green technological innovation and regional coordinated development.