A constant rate of change in harvest index (dHI/dt = k) has recently b
een incorporated into several crop simulation models, so that final gr
ain yield can be calculated from final biomass and the duration of gra
in growth. Implicit is the assumption that dHI/dt is conservative acro
ss treatments and environments. This assumption was tested using data
from five experiments grown in the United Kingdom (1973, 1978, 1994) a
nd New Zealand (1992, 1993). The experiments included commercial sprin
g and winter wheat cultivars introduced during the last 100 years and
nitrogen, irrigation, sowing date, temperature and CO2 treatments. In
all cases, the time course of harvest index (HI) had an initial lag ph
ase, a linear phase and a maturation phase. The linear phase was stabl
e in field-grown crops, except for a reduction in slope after lodging
in some crops. Values for dHI/dt, taken as the slope of the linear pha
se, varied with variety and available nitrogen, were stable for a give
n variety among years, and were unaffected by water stress. Variation
in dHI/dt among varieties was independent of their year of introductio
n, although those with the Rht2 semi-dwarfing gene generally achieved
a higher final HI due to a reduced lag phase. Differences in the durat
ion of the linear phase also caused differences in the final HI after
drought. The upper and lower limits of dHI/dt for field-grown crops we
re 1.37 and 0.64% d(-1) but, under normal fertility conditions, the va
riation was between 0.90 and 1.19% d(-1). Results indicated that dHI/d
t could provide an effective semi-empirical relationship for predictin
g grain yield in simulation models. The consistent, linear nature of t
his relationship suggests a physiological maximum for dHI/dt, for a gi
ven species and variety. It may be possible to exploit varietal differ
ences in dHI/dt, and in the lag phase, for yield improvement.