Multi-machine scheduling, that is, the assigment of jobs to machines such that certain performance demands like cost and time effectiveness are fulfilled, is a ubiquitous and complex activity in everyday life. This paper presents an approach to multi-machine scheduling that follows the multi-agent learning paradigm known from the field of Distributed Artificial Intelligence. According to this approach the machines collectively and as a whole learn and iteratively refine appropriate schedules. The major characteristic of this approach is that learning is distributed over several machines, and that the individual machines carry out their learning activities in a parallel and asynchronous way.