This work presents a novel nonlinear direct adaptive data cloud based fuzzy controller design method for a general class of NARMAX systems employing the evolving fuzzy approximation. In the proposed fuzzy controller, dynamic approximation capability is performed by the combined action of structure configuration and parameter adjustment using multiple strategies of recruiting, pruning and updating the data clouds, and self-tuning the consequent parameters. The normalization distance between two local densities is employed to measure the familiarity of each two data clouds, which is the criteria to trigger the pruning of redundant data cloud to further avoid the overlap or conflict. The simulation results demonstrate low computation cost and favorable tracking performance.