The response of eight long-duration pigeonpea [Cajanus cajan (L) Millsp.] genotypes to irrigation was studied at Gwalior in Central India during the 1990–91, 1991–92 and 1992–93 growing seasons on an Inceptisol. The crop was grown at two spacings as it was expected that crop density could interact with the crop's ability to extract soil moisture. The irrigation treatment received furrow irrigation four times during the 1990–91 and 1992–93 seasons and twice during the 1991–92 season. Grain yields of all genotypes were 11 % higher when planted at higher density than at low density. There was a differential variation in yield and harvest index among genotypes due to season but not due to spacing and irrigation suggesting the validity of the present approach of testing genotypes under optimum conditions. Grain yield declined by 21 % from the 1990 to 1992 season. The decline was > 1 t/ha in some cultivars (ICPL 366, GW3), and between 0·5 and 1·0 t/ha in others (NP [WR] 15, ICP 87143 and ICPL 84072). In others (Bahar, ICP 9174, ICP 8860) the yield fluctuation was < 0·5 t/ha. The genotypes' mean yields were as high as 2·7 t/ha for ICPL 87143, ICPL 84072 and ICPL 366. There was a significant reduction in both grain yield (16 %), and also above-ground plant dry mass (18 %) due to soil moisture limitation in the unirrigated treatment. Both the above-ground plant dry mass and grain yields were significantly more at high plant density than at lower plant density especially with irrigation. The genotypes were found to differ in their response to production environment (irrigation, spacing and to the undefined differences of the 3 years). Genotypic variation in yield within a production environment was found to vary in relation to changes in harvest index and across environment (irrigation, seasons) due to variation in total dry matter production. A lack of negative relationship between the total dry matter and harvest index suggests the possibility of optimizing both for obtaining higher yield from long-duration genotypes.