Abstract A multi-level allocation problem consists in determining the position and size of spaces on several levels according to the user's requirements and constraints. How levels are connected and how vertical circulation objects are treated may limit the practitioner's preferences or reduce the number of solutions. Literature review has shown that in the multi-level allocation problem very often stairs and elevators are considered as fixed and rigid objects during the search process and sometimes even neglected. This paper presents an approach to deal with multi-levels based on a hybrid evolutionary technique where stairs and elevators are parametric objects, which interact with other spaces during the search process. Three case studies are presented, where the algorithm is tested on handling large and complex problems, how concurrent vertical circulation objects interact, and how several levels are stacked within a building boundary. The algorithm shows to be capable of generating coherent multi-level floor plans.