We interpret galaxy counts in the B and K bands, and the redshift dist
ribution of faint galaxies, using a model for galaxy formation in a hi
erarchical universe. We first study how variation of each of the free
parameters in our model affects its observational predictions. These p
arameters control processes such as star formation, supernova feedback
and merging, and determine the masses and luminosities of the galaxie
s that populate the evolving distribution of dark matter haloes. We fi
nd that more efficient merging results in a galaxy luminosity function
with a lower amplitude at the faint end at all epochs. This causes th
e slope of the faint counts to decrease, and the redshift distribution
to shift to higher values of z. We then look at the influence of the
parameters that control star formation and feedback in dark matter hal
oes of differing circular velocity. We find that we can steepen the co
unts by allowing more stars to form in low-circular-velocity haloes. I
n order to avoid producing too many low-luminosity galaxies at the pre
sent epoch, star formation must stop before z = 0, allowing the galaxy
to fade substantially in the B band. The evolution of the satellite g
alaxies in our models provides a natural way of explaining this phenom
enon. Finally, we show that the choice of initial mass function does n
ot influence our results very substantially. We present two models whi
ch can fit most of the observational data: (a) a cold dark matter mode
l in which star formation is suppressed in low-circular-velocity haloe
s until they are accreted into larger systems, and (b) a mixed dark ma
tter model. Neither model fits perfectly. The amplitude of the field-g
alaxy luminosity function is too high at the faint end, and we are una
ble to account for the reddest objects seen at K magnitudes between 16
and 20. Given the uncertainties in both the observational data and th
e details of our modelling, our results lead us to conclude that neith
er unphysically high merging rates nor new populations of objects are
needed to explain the observations in the standard framework of hierar
chical clustering in an OMEGA0 = 1 universe.