We study the biasing relation between dark matter haloes or galaxies and th
e underlying mass distribution, using cosmological N-body simulations in wh
ich galaxies are modelled via semi-analytic recipes. The non-linear, stocha
stic biasing is quantified in terms of the mean biasing function and the sc
atter about it as a function of time, scale and object properties. The bias
ing of galaxies and haloes shows a general similarity and a characteristic
shape, with no galaxies in deep voids and a steep slope in moderately under
dense regions. At a comoving scale of similar to8 h(-1) Mpc, the non-linear
ity in the biasing relation is typically less than or similar to 10 per cen
t and the stochasticity is a few tens of per cent, corresponding to similar
to 30 per cent variations in the cosmological parameter beta=Omega (0.6)/b
. Biasing depends weakly on halo mass, galaxy luminosity, and scale. The ob
served trend with luminosity is reproduced when dust extinction is included
. The time evolution is rapid, with the mean biasing larger by a factor of
a few at z similar to3 compared with z=0, and with a minimum for the non-li
nearity and stochasticity at an intermediate redshift. Biasing today is a w
eak function of the cosmological model, reflecting the weak dependence on t
he power-spectrum shape, but the time evolution is more cosmology-dependent
, reflecting the effect of the growth rate. We provide predictions for the
relative biasing of galaxies of different type and colour, to be compared w
ith upcoming large redshift surveys. Analytic models in which the number of
objects is conserved underestimate the evolution of biasing, while models
that explicitly account for merging provide a good description of the biasi
ng of haloes and its evolution, suggesting that merging is a crucial elemen
t in the evolution of biasing.