Television research is increasingly dominated by a need to predict ratings accurately. Prognoses of future ratings are of crucial importance for media planners producing or scheduling programmes, and selling or buying commercial time. Forecasts are made by advertising sales houses and media buyers by means of a combination of instinct, experience, luck and elaborate models. To assist anyone forecasting ratings in the Netherlands, Intomart developed an entirely automated model to predict television ratings. Each month this model computes forecasts for all programmes on the five main channels for 17 target groups. These forecasts are calculated within a single day using new programme schedules and historical databases with, as input, people meter data. After a pilot phase, the system became fully operational as of October 1992 exclusively for the media buying organization Initiative Media. Since then, the model has been developed further in co-operation with Initiative Media who provided valuable support as well as stimulating criticism. Since April 1994 this forecasting model is also available to other operators in the industry. This paper will explain the theoretical background and technical aspects of the main determinants of this model. Additionally, the results will be evaluated in terms of accuracy and practical value for the industry.