Landmarks can capture the features of the solution space of planning tasks precisely. In this paper, a decomposed planning method guided by landmark information is proposed. The method executes an enforced hill-climbing procedure guided by the landmark-counting heuristic towards the goal, searching for a plan along with completions of the landmarks. Globally the hill-climbing procedure achieves the landmarks one after another. A decrease in the landmark counting heuristic estimation causes task decomposition and whenever the search encounters a state with a lower estimation, a hill-climbing fragment is extracted. Such "search-extract" procedure is repeated until the estimation of the landmark counting heuristic decreases to zero eventually, and then all the extracted fragments are connected into the final plan. Experiment results show that the decomposed planning method guided by landmark-counting heuristic makes use of the landmark information in a more flexible way, usually cutting down the search space dramatically and find the plan much faster.