The effect of natural selection on linked sites has been suggested to be a major determinant of genetic diversity. While it is in principle possible to estimate this effect from genome sequence data, interactions between selection, demography and inbreeding are expected to make inference less reliable. Here we investigate whether the genome-wide reduction in diversity due to background selection (B) can be accurately estimated when populations are at demographic non-equilibrium and/or reproduce by partial self-fertilisation. We show that the classic-BGS model is surprisingly robust to both demographic non-equilibrium and low rates of selfing, although both processes do lead to biased estimation of the distribution of fitness effects (DFE) of deleterious mutations. A high rate of selfing leads to poor estimation of both B and DFE parameters. We propose an alternative inference approach where background selection, demography and partial selfing are jointly estimated from windowed site frequency spectra. This approach, while approximate in nature, resolves most of the bias observed under the classic-BGS model and can also generate estimates of past demography that account for the effect of background selection and partial selfing. We apply the approach to genome sequence data from Capsella grandiflora and C. orientalis, which have contrasting mating systems and display a forty-fold difference in nucleotide diversity. Our results suggest that background selection has a weak effect on levels of genetic diversity in the outcrosser C. grandiflora (B=0.89) and a more substantial effect in the predominantly selfing species C. orientalis (B=0.44), but that background selection alone cannot explain their disparity in genetic diversity.