The mechanistic model KOSI was developed to predict the weekly fresh weight harvest of cucumber fruits and their quality. The model consists of modules for greenhouse climate, greenhouse light transmission, light interception by the crop, leaf and canopy photosynthesis, assimilate partitioning, dry matter production, fruit growth, fruit dry matter content and fruit harvest. The minimum data needed by KOSI for harvest prediction are date of planting of the crop and date scheduled for the last harvest. When only this minimum data set is used, calculations are based on long-term average data on weekly global radiation and temperature outside the greenhouse. Instead of using long-term average weather data, predicted or measured weather data can be provided as input. Model predictions can be improved by providing more input parameters, such as temperature set-point and CO2 concentration of the greenhouse air, plant density, fruit pruning, frequency of fruit harvesting and transmissivity of the greenhouse. Model output are weekly harvest of the total fruit fresh weight, fruit number, the fresh weight and age of individual fruits and the percentage of second class fruits. The model was validated by comparing simulation results with production data of 10 commercial growers in 1996 and 14 growers in 1997 (January - May). Even when only the date of planting of the crop and date scheduled for the last harvest were used as input to the model, the weekly harvest of total fresh weight averaged over all growers was simulated well by the model. The average error of the weekly prediction of the fresh weight yield was 14.9°while the error of the annual yield was 2.8␒n 1996. The predicted average fruit size and percentage of second class fruits corresponded reasonably well with growers' data, showing average weekly errors of 6.5 nd 5.3°respectively.