Summary There is ample evidence of the efficacy of cytological screening in the prevention of cervical cancer but disagreement on the form which screening programmes should take. Simulation models have been used as a convenient and rapid method of exploring the outcome of different screening policies and of demonstrating the importance and interrelationships of the variables concerned. However, most such models are either too abstract or too simplistic to be of practical value in planning screening programmes.A model is described which reproduces demographic events in a female population (that of England and Wales) over a 30 year period, and onto this superimposes the natural history of cervical carcinoma, using data derived from published studies. A microsimulation approach -each individual in the population being retained as a unit -allows factors such as disease onset and screening uptake to be dependent upon personal characteristics and past events. Screening can be offered as part of a routine programme, or incidentally -for example during pregnancy or hospital attendance.The model allows quantitative evaluation of the complex patterns of screening that are actually observed and the relative importance of the different components of such screening programmes. Assumptions about natural history can thus be studied.Cervical cytology screening is an effective means of reducing the incidence of invasive cancer of the cervix (Guzick, 1978;. However, the decision to introduce a screening programme into a population will depend upon considerations of the degree of benefit obtainable in relation to the resources deployed and this ratio will vary considerably with the actual programme design (who is to be examined, where, and how often). The utility of randomised trials to evaluate complex programmes involving repeated examinations and follow-up of large numbers of subjects is limited by the time and expense involved. It is possible to investigate only a limited number of screening schedules and the methodology used may be deemed obsolete or inappropriate by the time results become available. Finally, the apparently obvious benefit to be gained from early diagnosis means that it is difficult to convince physicians and their patients of the need for a controlled trial, so that they may be judged unethical, or the control group receive screening outside the trial.Health care models are a convenient way of predicting the results of different interventions, given varying assumptions about the relationships involved. Their advantages have been summarised by Eddy & Schwartz (1982 Coppleson & Brown (1975) is the simplest, a Markov process using just 4 states (normal, dysplasia, carcinoma in situ (c.i.s.) and invasive cancer) and a set of transition rates between them compatible with observed cross-sectional data on incidence and prevalence of the preclinical states (Bibbo et al., 1971), and cumulative incidence of clinical cancer (Cutler & Young, 1975 An identical approach using the same model has recently been published b...