The widespread use of UAV (Unmanned Aerial Vehicle) has made trajectory planning become a hot research area．Although there have been numerous UAV trajectory planning methods so far，it will be difficult to effectively apply in trajectory planning problems under complicated conditions on account of oversimplified planning models or inefficient planning methods．Therefore，an improved RRT-Connect (Rapidly-exploring Random Tree Connect) algorithm for this problem is put forward．Firstly，a mathematical model of UAV 3D trajectory planning problem is established．Then，when solving aforementioned model by adopting classical RRT-Connect algorithm，6 kinds of improved strategies for generating random nodes are put forward．Finally，the improved RRT-Connect algorithm are applied to UAV 3D trajectory planning problem under threat of thunderstorm．Compared with using classical RRT algorithm and classical RRT-Connect algorithm to solve the same problem，experiment results show that the improved RRT-Connect algorithm can efficiently generate feasible UAV trajectory．