We study the uncertainties in parton distributions, determined in global fits to deep inelastic and related hard scattering data, due to so-called theoretical errors. Amongst these, we include potential errors due to the change of perturbative order (NLO→NNLO), ln(1/x) and ln(1 − x) effects, absorptive corrections and higher-twist contributions. We investigate these uncertainties both by including explicit corrections to our standard global analysis and by examining the sensitivity to changes of the x,Q2,W2 cuts on the data that are fitted. In this way we expose those kinematic regions where the conventional DGLAP description is inadequate. As a consequence we obtain a set of NLO, and of NNLO, conservative partons where the data are fully consistent with DGLAP evolution, but over a restricted kinematic domain. We also examine the potential effects of such issues as the choice of input parameterization, heavy target corrections, assumptions about the strange quark sea and isospin violation. Hence we are able to compare the theoretical errors with those uncertainties due to errors on the experimental measurements, which we studied previously. We use W and Higgs boson production at the Tevatron and the LHC as explicit examples of the uncertainties arising from parton distributions. For many observables the theoretical error is dominant, but for the cross section for W production at the Tevatron both the theoretical and experimental uncertainties are small, and hence the NNLO prediction may serve as a valuable luminosity monitor.