Many areas of world, particular those where agriculture is largely pra
cticed by resource-poor farmers with little or no use of external inpu
ts, have not benefitted from the spectacular yield increases achieved
by the combination of modern breeding technologies and use of inputs.
The paper argues that because breeding is mostly conducted in presence
of high inputs, it has systematically missed the opportunity to explo
it genetic differences at low levels of inputs. Many studies show that
these differences do exist, particularly in the case of fertilizers,
and that these differences can only be identified is selection is cond
ucted under the target level of inputs. Although this was predicted by
theory more than 40 years ago, and has been supported by a large body
of experimental data, very few breeders select in sub-optimal or stre
ss conditions. The most common justification is the high environmental
variation, and hence the lower heritability expected in low input con
ditions. While this is not supported by experimental evidence, the pap
er shows that in the case of a typical crop grown in low-input and cli
matically marginal conditions such as barley, genetic gains are possib
le by using locally adapted germplasm and by selecting in the target e
nvironment. Similar conclusions, in relation to the use of a low-input
selection environment, have been reached recently in maize. It is con
cluded that the best avenue to a sustainable increase of agricultural
production in low-input agricultural systems is through locally based
breeding programs.