Generalizing a well known theorem for finite matroids, we prove that for every (infinite) connected matroid M there is a unique tree T such that the nodes of T correspond to minors of M that are either 3-connected or circuits or cocircuits, and the edges of T correspond to certain nested 2-separations of M . These decompositions are invariant under duality.(ii) This decomposition is unique in the sense that for any two such treedecompositions, (T, R) and (T , R ) say, there is an isomorphism v → v between the trees such thatSince k-separations of a matroid M are also k-separations of its dual M * [16], a tree-decomposition of M is also one of M * , with the same adhesion. Moreover, the torsos corresponding to a given tree node are duals of each other: Theorem 1.2. Every tree-decomposition (T, R) of a connected matroid M is also a tree-decomposition of its dual M * . If (T, R) has uniform adhesion 2 for M , it has uniform adhesion 2 also for M * , and (M v ) * = (M * ) v for all v ∈ T . In particular, M and M * have the same unique irredundant tree-decomposition.The notation we use in this paper is as follows. Axiom systems for infinite matroids can be found in [15]. For other terminology we follow Oxley [22], or [21] for graphs. The letter M always denotes a matroid. Its ground set, set of bases, and set of circuits will be denoted by E(M ), B(M ) and C(M ), respectively. Given S ⊆ E(M ), we let M |S and M/S denote the restriction of M to S and the contraction of S in M , respectively, and write S := E(M ) \ S. The dual matroid of M is denoted by M * .