The forced expiratory volume in one second (FEV1) is routinely used in epidemiologic studies of lung function to assess the presence and severity of obstructive airways disease. Normative prediction equations developed using data from healthy, asymptomatic individuals may then be used both in a clinical setting and to adjust comparisons among risk subgroups for known demographic differences. Unfortunately no concensus has yet developed as to how best to model lung function data. This paper addresses this issue in a systematic manner using data derived from two cohorts followed over a period of 9-11 years. We compare a variety of cross-sectional and longitudinal models for FEV1, show how they may be expressed as members of a larger class of general linear models, and discuss goodness-of-fit procedures for comparing them. We found little objective evidence for discriminating among these models; only those fit to FEV1/ht3 performed poorly. We argue on subjective grounds for the use of models based on FEV1 as a function of age, height and their interactions.