Adjustment for multiple confounding is usually done by applying some kind of regression model. This requires assumptions about the type of relationship between the exposure of interest and outcome. It has been suggested to use confounder scores within strata of which the relationship can be scrutinised without such assumptions. Adjusting continuous outcomes for confounding factors, like birth weight for gestational age and lung function measures for age, is sometimes done by using the ratio of the observed outcome to an outcome expected from a reference series external to the study, with a concomitant risk of introducing new confounding. In some cases, the ideas from both these approaches can, slightly modified, be successfully combined. By using an internal rather than an external reference series, i.e. a group within the data with constant exposure, expected values for outcome can be derived as the fitted values from some model, appropriate for the purpose, depicting outcome conditional on all the potential confounders that are registered and for which control is desired. The relationship between the exposure of interest and the outcome, now represented by the fractional departure for each individual from the value expected conditional on her/his confounder status, can be scrutinised without any assumptions about the form of this relationship but, contrary to the confounder score approach, on the basis of the entire data set. This method also provides a way of presenting the relationship of interest, adjusted for confounding, that is easier to understand than the traditional regression approach. The impact of alcohol intake during pregnancy on birth weight will be given as an example.