In a test of three computer models to simulate group decisions, data were used from 31 American and Austrian groups on a total of 307 trials. The task for each group was to predict a series of answers of an unknown subject on a value-orientation questionnaire, after being given a sample of his typical responses. The first model, used the mean of the individual opinions as a simulation of the group judgment, simulated exactly over half of the trials. The simulation was improved in model two, which also used the mean, when individual opinions were weighted according to their total participation in the discussion and "learning" was added. This was especially true when the cutting points were changed so that the mean would represent more extreme opinions. The best simulation occurred with model three, which used the median of the individual opinions as a simulation of the group decision. Using this model, over 75% of the trials were accurately simulated. (Author)