A multi-year (2004-2009) field survey of on-farm sowing practices in 10 villages located in south-west Niger close to Niamey, is analysed to investigate the relationships (i) between rainfall and the sowing date of pearl millet and the risk of sowing failure, (ii) between sowing and meteorological/agro-climatic onset dates, (iii) between sowing/onset dates, and simulated and observed yield/biomass at the end of the season. Even if some villages sow without any synchronous or anterior rainfall, most parcels (73% out of the 1551 available cases) are sown during and just after a 2-day wet spell receiving at least 10 mm. In fact, there is a strong correlation (r = 0.82-0.95 depending on onset definition) between the spatial averages of onset and of sowing dates. Most of the failed sowings (?22% of total sowings) are related to dry spells lasting at least 7 days after an initial 2-day wet spell receiving less than 10 mm. Simulations with the "Système d'Analyse Régionale des Risques Agronomiques - version Habillée" crop model show that the ideal sowing date, retrospectively computed as the one maximizing simulated yield, is on average about 6 days later than the observed one. Despite the large inter-village variance and the relatively weak inter-annual signal in onset dates and seasonal amounts, there is a tendency for weaker yields and especially weaker amounts of biomass for late onset. But crop simulations show that sowing very early, as for example during or just after the first wet spell when at least 90% of rainfall stations receive simultaneously at least 1 mm in two consecutive days (i.e. meteorological onset), does not necessarily maximize simulated yield because of the high risk of long-lasting post-onset dry spells. The farmers' strategy, that is sowing their field during or just after the first significant wet spell, is combined with the use of photoperiodic varieties to provide the best-suited response to the temporal and spatial variability of onset of the r