We used a computer program to manage marker data in a recombinant inbred li
ne population. The objective was to select pairs of inbred lines to be inte
rcrossed, in order to cumulate all favourable alleles, either with additive
effects or with interactive effects. The population size required to have
95% chance of obtaining the best line from a given cross is computed, takin
g into account the number of QTLs and the probability that no recombination
event occurs in any QTL confidence intervals. It is shown that the accurac
y of QTL location greatly affects selection efficiency and that a recurrent
selection scheme is highly preferable for pyramiding many QTLs. Dough stre
ngth (W) was chosen to illustrate the use of this method. In a population o
f 187 DH lines from the cross Courtot x Chinese Spring, W was found to be c
ontrolled by 8 additive QTLs and two pairs of interactive QTLs. None of the
187 DH lines possess all favourable alleles or combination of alleles. The
best possible line could be produced by a single cycle of crossing followe
d by line extraction, but several thousand lines needed to be produced to a
chieve this goal. Alternatively, a recurrent crossing scheme allows to obta
in the best line in two generations, and requires less than 150 lines per c
ycle.