Abstract. Hyper-minimization aims to reduce the size of the representation of a language beyond the limits imposed by classical minimization. To this end, the hyper-minimal representation can represent a language that has a nite dierence to the original language. The rst hyperminimization algorithm is presented for (bottom-up) deterministic tree automata, which represent the recognizable tree languages. It runs in time O( mn), where is the maximal rank of the input symbols, m is the number of transitions, and n is the number of states of the input tree automaton. We generalize hyper-minimization to deterministic tree automata (dta) [5,6], which have applications in XML processing [10] and natural language processing [11]. We faithfully generalize the existing denitions from dfa to dta.