This paper presents a novel approach to an artificial immune system, which uses the rough set theory to improve its classification ability under uncertainty in data. The proposed approach is mainly based on a negative selection algorithm and is suitable to solve problems where the knowledge of non-self is scarce and noisy. The rough set theory is used to deal with uncertainty in data and to obtain rule sets necessary to specify both, self and non-self classes. The proposed artificial immune system is implemented with rough valued lymphocytes, which emulate the approximate binding performed by natural immune systems.