Global monsoon precipitation plays a crucial role in the local social economy and global large-scale circulation and energy cycle. Using the decadal prediction output for 1960-2015 from ENSEMBLES Stream 2, the decadal hindcast skill of climate models on global land monsoon precipitation and the potential source of predictability were examined in this paper. It is found that the decadal variation of global and southern hemispheric land monsoon precipitation is not well hindcasted by ENSEMBLES. However, the Northern Hemispheric land Summer Monsoon (NHSM) precipitation in hindcast is well predicted, including the observed downward trend from 1960 to the late 1970s and upward trend since the 1990s. The main deficiency is that the minimum NHSM precipitation occured in mid-1970s, which is 10-year earlier than the observation, leading to poor prediction of NHSM precipitation from the mid-1980s to early 1990s. Mega-ENSO and Atlantic Multi-decadal Oscillation (AMO) are the two main factored that modulate the decadal variation of NHSM precipitation. The result shows that the relationships of NHSM precipitation with mega-ENSO and AMO in ENSENBLES are higher than the observation. The climate models well predicted the increase from 1960 to the late 1970s and decrease trend since the 1990s of mega-ENSO and AMO. It is the primary source of the prediction skill on NHSM changes during the two periods. Although AMO is well predicted by ENSEMBLES (highest correlation coefficient with observation is 0.85), the prediction skill of mega-ENSO is limited, leading to poor performance in predicting NHSM precipitation from the mid-1980s to early 1990s. Thus, improving the prediction of mega-ENSO can be seen as one important method of better decadal prediction of NHSM precipitation.