Determining the number of kanban in a pull production is essential for a good system performance. While this is simple in a static case, it becomes much more challenging if the system is subject to variations, as all practical systems are. The paper presents a novel boundary case approach to determining the number of kanban subject to a desired delivery performance. The approach is compared with the well-known Toyota formula. An extensive simulation-based approach verification for different coefficients of variations for the replenishment time and the customer takt was used to verify this boundary case approach also in comparison with the Toyota formula. The presented approach handles variation much better and determines a much more suitable number of kanban. This approach can also be adapted to practical everyday use in industry with little additional effort and with only a limited understanding of the underlying randomness of the key variables.