Relationship between start of the rainy season and farmer's planting date have been analysed at local scale through (i) in situ daily rainfall records and (ii) agronomical measurement provided by on-farm survey at plot scale for 10 villages network located on the mesoscale AMMA-CATCH Niger site during the 2004-2007 rainy seasons. Many classical empirical methods of monsoon onset dectection (e.g. Benoit, Kowal, Sivakumar, Ati, Marteau, Balme, Sultan) and numerical methods based on crop model SARRAH have been compared to the observed farmer's sowing date. Results show that SARRAH and Balme mean onset date are close to sowing date. But spatial and temporal variability of the sowing date is not correctly reproduced by any start of the rainy season. Nevertheless, almost 50% of the successful observed sowing date are in phase with the start of the rainy season defined by Sivakumar, Balme and SARRAH whereas about 40% of sowing are in advance comparing to Sivakumar or SARRAH onset date. Farmers' management of the sowing date is systematically based on risk strategy in order to ensure a maximum yield for most of the year. Thus, 23% of first sowing, usually earlier than the start of the rainy season, have failed and require re-sowing due to the intermittent dry spells and low rainfall intensity after sowing. Indeed, it causes a strong crop water stress during germination and panicle initiation. Otherwise, the start of the rainy season and sowing date does not seem to be the main factor of the yield variability, because these parameters explain only 10% and 16% respectively of the yield variation. However, mean yield appears to be affected by the delay of the start of the rainy season and/or latter sowing date which reduces grain yield of about 30% than ealier start of the rainy season or sowing.