In this work the disturbance decoupling problem in decentralized linear systems (DDPD) is formulated and solved without making any assumption on the kind of decentralized pattern of information in the system. Necessary and sufficient conditions are given under which disturbance decoupling is possible by applying local control laws formed by a combination of measured disturbance feedforward and non-dynamic feedback of state or measurement. For the solution of DDPD, the new geometric concepts ofk—(A, Bi, 𝓈i)- andk-(CiA, Bi)-invariant subspaces are developed. They extend the(A, B)- and(C, A, B)-invariance respectively to decentralize systems. A geometric description of the maximal elements of the class of these subspaces that are candidates for generating a solution to various DDPDs considered are also developed. Numerical examples are given.