Abstract: Trust region methods are a class of numerical methods for optimization. They compute a trial step by solving a trust region sub-problem where a model function is minimized within a trust region. In this paper, we review recent results on non-monotone trust region methods for unconstrained optimization problems. Generally, non-monotone trust region algorithms with non-monotone technique are more effective than the traditional ones, especially when coping with some extreme nonlinear optimization problems. Results on trust region sub-problems and regularization methods are also discussed.