Many-objective optimization problems (MaOPs) are vital and challenging in real-world applications. Existing evolutionary algorithms mostly produce an approximate Pareto-optimal set using new dominance relations, dimensionality reduction, objective decomposition, and set-based evolution. In this paper, we propose a mutation operator guided by preferred regions to improve an existing setbased evolutionary many-objective optimization algorithm that integrates preferences. In the proposed mutation operator, optimal solutions in a preferred region are first chosen to form a reference set; then for each solution within the individual to be mutated, an optimal solution from the reference set is specified as its reference point; finally, the solution is mutated towards the preferred region via an adaptive Gaussian disturbance to accelerate the evolution, and thus an approximate Pareto-optimal set with high performances is obtained. We apply the proposed method to 21 instances of seven benchmark MaOPs, and the experimental results empirically demonstrate its superiority.