The interannual variability of the onset and cessation dates of the rainy season (RS) in Senegal is analyzed over the 43 yr period 1950-1992, using daily rainfall data for 34 stations. The use of principal component analysis, based on rainfall only, is explored to identify aggregate, regional indexes for the onset and cessation of the rains. The minimum and maximum values of the cumulative scores of principal component 1, for each year, are used to locate the onset and cessation dates, respectively. Very distinct spatial rainfall patterns are found before and after the onset/cessation dates. Mean dates compare favorably with those based on other definitions, though our method is not meant to provide dates with a local, rigorous agronomic significance. Over 1950-1992, the time-series for the cessation shows a significant trend towards earlier dates, with an abrupt shift occurring around 1970. There is also a small trend for a delayed onset of the rains. However, the interannual variability of onset is greater than that of cessation dates. Correlation between onset, cessation, duration and total rainfall amounts is generally significant, but not as high as expected. 1988 and 1989 are good illustrations, with nearly identical rainfall amounts but an onset delayed by 1.5 mo in 1988 as compared to 1989. Graphs obtained using this method provide other interesting information about the RS, such as the general performance of the RS in terms of total rainfall amount. Pronounced dry spells are easily located: 29 cases were found between 1950 and 1992, concentrated in the early part of the RS (June-July). 'False starts' of the rains can also be monitored; they were reported in 16 out of the 43 years. Frequencies of dry spells and false starts do not exhibit any particular long-term trends.
