Intuitively, the security of a steganographic communication between two principals lies in the inability of an eavesdropper to distinguish cover-objects from stego-objects, that is objects which contain secret messages. A system should be already considered insecure, if an eavesdropper can suspect the presence of secret communication. Several definitions of steganographic security were proposed in the literature. However, they all consider only perfectly secure steganographic systems, where even a computationally unbounded observer cannot detect the presence of a secret message exchange. Second, it might be difficult to construct secure schemes usable in practice following these definitions. Third, they all require the knowledge of the probability distribution of normal covers; although it might be possible in certain cases to compute this probability, it will in general be infeasible to obtain. In this paper, we propose a novel approach for defining security in steganographic systems. This definition relies on a probabilistic game between the attacker and a judge. Given the ability to observe the normal communication process and the steganographic system, the attacker has to decide whether a specific object (given to him by a judge) is in fact a plain cover or a stego-object. We discuss the applicability of this new definition and pose the open problem of constructing provably secure steganographic systems.