In this paper we analyse the deep number counts problem, taking account of new observational and theoretical developments. First we show that the new Bruzual and Charlot (1993) models allow a new class of spiral dominated luminosity evolution (LE) model where significant amounts of the luminosity evolution needed to fit faint count data are due to spiral rather than early-type galaxies. Second we show that the inclusion of dust may be a vital ingredient for obtaining fits with any LE model. Third we compare the quality of fit of both the spiral and early-type LE models, including dust, for a wide variety of observational data. We find that parameters can be found for both LE models which allow a good fit to all data with the exception of the faintest B>25 counts in the case of q0=0.5 cosmologies, where some luminosity dependent evolution may be needed (see also Metcalfe et al 1995). Otherwise both these classes of LE model, with the inclusion of dust, provide an excellent foundation for understanding the B<25 galaxy counts and galaxy counts and redshift distributions in a variety of other wavebands.