Brother trees were proposed, by some of the authors of the present article, as a balanced tree scheme for the implementation of dictionaries (data structures which support the operations of membership testing, deletion and insertion). A brother tree is defined as a rooted, directed tree satisfying the following three conditions: (1) each internal node has outdegree 1 or 2, (2) each unary node has a binary brother (a node is k-ary if it has outdegree k), and (3) all root-to-leaf paths have the same length. Two possible ways of storing keys in the nodes of the tree lead naturally to 1-2 brother trees and brother leaf-search trees. The authors discuss three cost measures for brother trees: node-visit cost, comparison-cost, and space-cost. They characterize 1-2 brother trees which are optimal with respect to the three cost measures and provide linear time algorithms for their construction. Optimality of brother leaf-search trees is discussed briefly.