Statisticians are too often satisfied by fitting data rather than investigating the process out of which the data arose. The assumptions on which they base their models may be quite unrealistic, and while it is true that a model should not be more complicated than necessary, neither should it be too simple. Ways of approaching several sets of data from different areas of clinical medicine are considered, and different attitudes to the purpose of modelling highlighted. The transition from smoothing data, through fitting curves, to modelling underlying processes is discussed.