Rainfall onset has a lot of implications on the sustainability of the socioeconomic activities in Nigeria. This study assesses the skills of CMA, ECMWF, and UKMO sub-seasonal-to-seasonal (S2S) models in predicting monsoon onset and its variability over Nigeria. It also investigates the global drivers modulating the variability and their teleconnections with rainfall onset anomaly. All the models, their ensemble members, and the observations were subjected to quantitative statistical analyses from 1998 to 2016. Results show that the three models are able to simulate the Northwards migration of the onset dates adequately with inherent biases and unique characteristics. They are also able to capture the evolution and variability of the global drivers modulating the monsoon onset. While CMA and the ECMWF models improve progressively towards the Sahel, the UKMO model performance is best over the Gulf of Guinea. In addition, despite the fairly poor performance of the models in predicting the variability of onset dates over the Gulf of Guinea and the Sahel, there is a considerable improvement in the correlation skill of the models over the Savannah. Furthermore, results show that only the ECMWF model was able to produce the strength of both the African Easterly Jets (AEJ) and the Tropical Easterly Jet (TEJ) in spatio-temporal mode. These are two of the crucial global drivers modulating the dynamics of West African monsoon. It was also found out that most global drivers, especially the Inter-tropical Discontinuity (ITD) and the Sea Surface Temperature (SST) over the Central Pacific, exhibit direct teleconnection with the onset anomaly. This direct relationship is shown to be strongest over both the Gulf of Guinea and the Sahel. Although the CMA model might have the least skill, it, however, showed that all the S2S models, despite the inherent biases, are able to predict rainfall onset over Nigeria, within the sub-seasonal timescale. Finally, the results show that improvements in multi-model ensembles are valuable added information able to significantly improve model performance.