Most of the literature estimating the gender wage gap (GWG) in the UK uses regression techniques adjusting for a range of covariates measured in adulthood, many of which are likely endogenous. In doing so, they typically find that about half the raw gap can be accounted for by gender differences in those conditioning variables. Using data for a birth cohort born in 1958 collected prospectively across ten survey sweeps, I depart from the literature in two ways. First, I condition solely on covariates measured in childhood to minimize potential biases arising from conditioning on endogenous variables. Second, I use a range of matching estimators to recover the GWG among ‘like’ men and women. I show that the raw GWG rises until cohort members enter their 40s, then falls but remains sizeable until age 63. In contrast to literature, when conditioning solely on childhood variables the covariate-adjusted GWG is similar to the raw GWG. This is the case at all ages regardless of the matching estimator I use and the set of conditioning childhood covariates. The implication is that among those born in 1958 women have lower earnings than men across the life-course which cannot be accounted for by their pre-labour market experiences.