This paper proposes a behavior selection algorithm for humanoid robots to perform complex tasks by defining four input (context) symbols and seven target (atom behavior) symbols. To employ the degree of consideration-based mechanism of thought (DoC-MoT) in the algorithm, the consideration degree for each input symbol is represented by the fuzzy measure and the knowledge link strengths between input and target symbols are represented by the partial evaluation values. Then, each atom behavior is globally evaluated by the fuzzy integral of partial evaluation values with respect to the fuzzy measure values and the atom behavior with the highest evaluation value is selected and activated.