Sorghum is known as a short day plant requiring short photoperiods (PP) to flower. However, in field experiments based on monthly plantings reported elsewhere, panicle initiation took place during long days as well as during short days even for highly photoperiod-sensitive varieties thought to be of the absolute type, or was particularly early for the sowings in September, well before days were shortest. In order to investigate these contradictory results, three tropical sorghum varieties of different photoperiod sensitivity were planted monthly in Mali and the dates of panicle initiation (PI) were recorded. The results indicate that the common concepts of a gradual (linear, or quantitative) response of photoperiod-induced phase (PIP) to PP at PI (PPi) (case of the variety Sariaso 10) or a threshold-type response (hyperbolic, or absolute; cases of CSM 335 and IRAT 174) could be reproduced for the sowing dates falling into the wet season (May-October), but did not describe adequately crop behaviour during the remaining months of the year. The same varieties were tested in controlled environments with four differential PP treatments composed of three different initial PP (12 h 15 min, 12 h 00 min and 11 h 45 min) and two opposite rates of day length change (± 1 min/day). Thus after 15 days, two pairs of treatments with opposite day length changes reached the same average PP (12 h 07 min and 11 h 52 min) and after 30 d, another pair reached a similar average PP (12 h 00 min). In 2001, 2 varieties reached PI earlier under decreasing days and the same average PP, but no such additional effect of the rate of change of PP was observed in 2002. Modelling exercises showed that field variation in the duration to PI was better explained with rate of change of PP than with absolute PP observed during PIP, and best with a combination of both factors in an additive model. It is concluded that in tropical sorghum, floral induction is strongly associated with a negative rate of cha