Structural variability is a fundamental property of biological systems. Fermentation processes, in particular, are often accompanied by physiological phenomena which have the characteristics of structural alterations. The conventional control approach does not have the competence to handle such processes, because it is based on the general assumption for a single-structure plant. In order to overcome this limitation, we consider under some conditions fermentation processes as variable structure plants and propose a two-level hierarchical scheme for their control. At the higher hierarchical level, which performs the organizing functions and operates in the structural space of the plant, the current plant structure is recognized as an element of a finite set of structures defined on the basis of ‘deep’ expert knowledge of the cell physiology and/or ‘shallow’ expert knowledge of the behavioral characteristics of the microbial population. The recognition is based on fuzzy decomposition of the structural space of the plant into several subspaces, mapped isomorphically into a respective set of rational control strategies. A set of expert rules is used for the actual synthesis of the recognition procedure, which is finally tuned by supervised learning. The adopted methodology at this hierarchical level provides the system with a kind of intelligence. To the lower hierarchical level, two distinct functions are assigned. The first one performs the global task of control of the plant structure and operates in its structural space. The second function has a local character and consists of control of the plant in a particular subspace of constant structure. It works in the state space of the plant. Under supervision of the higher level, the control strategy relevant to the current plant structure is picked out from a predefined pool and the corresponding control action is calculated. Extensive use of expert knowledge in the conceptual formulation of proper control strategies is assumed. Compared with the conventional single-structure approach, the variable structure concept results in rational and physiologically motivated control of fermentation processes. Instead of the development of a single but complicated process model and control structure, construction of several simple control structures, often in a ‘model-less’ form, is required. This method naturally incorporates and theoretizes the widely accepted policy of artificial induction of structural transformations in the plant, which is essential for enhanced productivity in many modern fermentation technologies. The proposed approach has been realized as a real-time software system, provided with enough flexibility to be used with different fermentation processes and equipment. This system is orientated to IBM personal computers and works under the QNX multi-tasking real-time operating system.