Based on a panel data of 78 counties in Guizhou from 2015 to 2020, this paper incorporates agricultural net carbon sink into expected output, uses the Malmquist-Luenberger index based on the SBM directional distance function to measure the agricultural eco-efficiency, and comprehensively applies the Dagum Gini coefficient decomposition method, kernel density estimation method, and β-convergence model to systematically analyze the spatial-temporal evolution characteristics of agricultural eco-efficiency in Guizhou. This study finds that the agricultural eco-efficiency of Guizhou and its five regions showed a fluctuating and increasing trend. The intra-reginal differences of each region showed an expanding trend, indicating that the imbalance of agricultural eco-efficiency in different counties within each region has become increasingly prominent. The inter-regional differences of the five regions showed a generally expanding trend, indicating that the agricultural eco-efficiency in different regions was gradually unbalanced. By decomposing the differences this paper also finds that he inter-regional differences had exceeded the intensity of trans-variation, which became the main source of differences. The kernel density distribution curves of Guizhou and its five regions showed a rightward trend during the observation period with different degrees of polarization trends. The whole province and the five regions basically supported absolute β-convergence, and after considering the different influencing factors, the agricultural eco-efficiency in different regions will gradually converge to their own steady-state level.