Agent-based simulation is increasingly used to study systems in many areas of business and science. Using agent-based simulation for prediction could be very valuable. However, these models usually have many parameters which are difficult to measure directly leading to uncertainty as to the best values to use. It is very important to consider the alternative feasible sets of parameter values, otherwise the modelling results could be misleading. A method is described to take account of this when making predictions and was applied to an agent-based word of mouth consumer model. The result was a wide prediction range indicating the limitations of using such a model for predictions. The method could also be applied in other situations.