This paper considers the changing spatial pattern of infant mortality in England and Wales over the last century for a constant area definition. The analysis proceeds at the level of registration districts as defined in the 1890s, generated using the British Isles Historical Geographical Information System. Infant mortality at this spatial definition is followed over four points in time, the most recent being the early 1990s. Using a variety of inequality measures, the analysis demonstrates a decline in relative small-area differentials in the chances of infant deaths up to 1958, but some resurgence since then. Regression methods are then used to evaluate the effect of available social variables, such as population density and housing conditions. The analysis incorporates adjustments for the sampling variability in what are often small counts of infant deaths by adopting pooling strength techniques. Urban-rural differentiation has followed the same evolution: it was most marked in the 1890s and reappears by the 1990s, after being reduced somewhat in the earliest years of the UK National Health Service. Differentials in infant mortality according to the housing quality and economic status of different registration districts also gain in importance in the twentieth century compared with the 1890s.