As an effective path planning approach, Potential Field Method has been widely used for Autonomous Underwater Vehicles (AUVs) in underwater probing projects. However, the complexity of the realistic environments (e.g. the three dimensional environments rather than two dimensional environments, and limitations of the sensors in AUVs) have limited most of the current potential field approaches, in which the approaches can only be adapted to theoretic environments such as 2D or static environments. A novel heuristic potential field approach (HPF) incorporating a heuristic obstacle avoidance method is proposed in this paper for AUVs path planning in three dimensional environments which have dynamic targets. The contributions of this paper are: (1) The approach is able to provide solutions for more realistic and difficult conditions (such as three dimensional unknown environments and dynamic targets) rather than hypothetic environments (flat 2D known static environments); (2) The approach results in less computation time while giving better trade-offs among simplicity, far-field accuracy, and computational cost. The performance of the HPF is compared with previous published Simulated Annealing (SA) and Genetic Algorithm (GA) based methods. They are analyzed in several environments. The performance of the heuristic potential field approach is demonstrated through case studies not only to be effective in obtaining the optimal solution but also to be more efficient in processing time for dynamic path planning.