Abstract. We consider the max-weight integral multicommodity flow problem in trees. In this problem we are given an edge-capacitated tree and weighted pairs of terminals, and the objective is to find a max-weight integral flow between terminal pairs subject to the capacities. This problem was shown to be APX-hard by Garg, Vazirani and Yannakakis [Algorithmica, 1997], and a 4-approximation was given by Chekuri, Mydlarz and Shepherd [ACM Trans. Alg., 2007]. Some special cases are known to be exactly solvable in polynomial time, including when the graph is a path or a star. First, when every edge has capacity at least µ ≥ 2, we use iterated LP relaxation to obtain an improved approximation ratio of min{3, 1 + 4/µ + 6/(µ 2 − µ)}. We show this ratio bounds the integrality gap of the natural LP relaxation. A complementary hardness result yields a 1 + Θ(1/µ) threshold of approximability (if P = NP). Second, we extend the range of instances for which exact solutions can be found efficiently. When the tree is a spider (i.e. if only one vertex has degree greater than 2) we give a polynomial-time algorithm to find an optimal solution, as well as a polyhedral description of the convex hull of all integral feasible solutions.