In clinical chemistry a linear-regression model may be used to determine reference intervals. A crucial point in this approach is the choice of variables to introduce into the model. In the present paper, we have applied a nonautomatic selection procedure, known as "element analysis," to a sample of 126 individuals from a small, ethnically homogeneous community in southern Italy. We investigated the effects of four independent variables--sex, age, weight, and alcohol consumption--on values for serum urea. Only sex and age proved to affect the urea values and were therefore introduced into the final model. This approach may be useful in determining reference intervals from observational studies when it is difficult to control a priori relevant factors. Moreover, variables may be selected not only on the basis of statistical criteria, but also according to biochemical and medical criteria.