This study addresses the need for cross-training in a cellular manufacturing environment. It is demonstrated that an effective cross-training situation results if workers and machines are connected, directly or indirectly, by task assignment decisions. The connections between workers and machines (i.e. the qualifications of workers) form,chains' that can be used to reallocate work from heavily loaded workers to less loaded workers. This provides the possibility of a balanced workload situation among workers, something that is desirable from a social as well as an economic viewpoint. Based on this insight, we have developed an integer programming (IP) model that can be used to select workers to be cross-trained for particular machines. The model may help in trade-offs between training costs and the workload balance among workers in a manufacturing cell. The workload balance indicates the usefulness of labor flexibility in a particular situation. A numerical example is presented to illustrate various elements and features of the model. It also provides further insight into the role of 'chaining' workers and machines. The industrial applicability of the model and directions for future research are also indicated. (c) 2004 Elsevier Ltd. All rights reserved.