A key to an optimal assortment of goods and pricing of individual items in a store is the knowledge about potential customer’s behaviour. In this paper we present the simulation of individual customers based on a multiagent system which models the important elements and external influences as single agents. An agent can be member of several agent groups which are represented as holons. We model each individual customer as an agent which behaves according the customer’s individual preferences. These preferences are extracted from real world data, such as customer cards, sales data and interviews. The customer’s shopping behaviour is represented in behaviour networks (Bayesian nets) which are stored in the customer agents’ knowledge bases. The behaviour of a representative group of customers induces the overall sales figures, which support decisions what to sell at which price. The presented concepts are based on ideas of Joachim Hertel from DACOS and Jörg Siekmann from the DFKI. They are implemented as a prototype, which provides, after further evaluation, the basis for a new and final system to be used by retailers.