Deep CCD imaging at the Isaac Newton Telescope provided a sample of ~7000 faint galaxies, with B-R colours, to faint limits of B~=25.5 mag and R~=24.5 mag. The number counts of these galaxies are consistent with the previous galaxy number counts of Metcalfe et al. and Tyson, showing at faint magnitudes an excess above non-evolving predictions, consisting of galaxies with moderately blue colours of 0.0<=B-R<=1.5. The number of redder (B-R<~1.5) galaxies did not exceed the non-evolving prediction. We calculate the angular correlation function, omega(theta, of these faint galaxies, and compare with the predictions of new models which take into account the observed dependence of galaxy clustering properties on morphology and luminosity. In agreement with previous studies (e.g., by Roche et al.), the omega(theta) amplitude of faint galaxies in our data set is found to fall well below the predictions of models in which galaxy clustering remains stable in proper coordinates and the redshift distribution N(z) has a non-evolving form, suggesting that the faint galaxy N(z) is more extended than this and includes a high proportion of z<~1 galaxies. We find that a pure luminosity evolution model, with a steep luminosity function for late-type galaxies, a significant increase with redshift in the star-forming activity of Sab and Sbc spirals, and mild dust-reddening, gives a good fit to both the galaxy number counts and the omega(theta) results, as obtained from this data set and from other published CCD surveys. However, the peak in the faint galaxy colour distribution is closer to B-R~=0.8 than to the bluer colour of B-R~=0.4 predicted by our model (the reason for this discrepancy remains unclear). The omega(theta) amplitude of the redder (B-R<~1.5) galaxies in our sample appears to be higher than that of the bluer (B-R<1.5) galaxies, in agreement with Roche et al. and Neuschaefer et al. We also find very little cross-correlation between red and blue galaxies. The lower omega(theta) amplitude of the bluer galaxies suggests that the steep fall in galaxy omega(theta) amplitudes at B<~23 is caused by the same blue galaxies that produce the excess in the number counts at these magnitudes. The colour dependence of omega(theta) is well fitted by our PLE model, in which, at B~25, almost all of the redder galaxies would lie at z<1, while the bluer galaxy subsample would consist of both dwarf late-type galaxies at low/moderate redshifts and evolving L~L* galaxies widely distributed in redshift from z~0.5 out to z~3, giving a much more extended N(z), and consequently a lower omega(theta) amplitude, for bluer galaxies. The stronger clustering of the redder galaxies suggests that galaxy clustering is approximately stable (epsilon=0) out to at least z~0.6, and therefore that the low omega(theta) amplitude of the full sample does not result simply from very rapid evolution of the clustering of all galaxies. These results may argue against models, such as merging-dominated models, in which the excess blue galaxies are confined to lower redshifts. The excess blue galaxies could only lie within a no-evolution N(z) if they belong to a separate population of dwarf starburst galaxies which is intrinsically very weak clustered. However, there is evidence (e.g., Cole et al.) that dwarf starburst galaxies are normally clustered. This would leave significant L* evolution, with a redshift distribution extending to z~3 at B~25, as the most plausible explanation of the low omega(theta) amplitude of B~25 galaxies.