Abstract: Fuzzy backward heuristic search algorithm FDA* is presented for fuzzy multistage decision problems.It is proved that with admissible heuristic evaluation function ,FDA*　algorithm is admissible and can always find out the best decision sequence with minimal cost.In terms of the fact that in general an admissible  is very difficult to design,the asymptotic learning algorithm Learning for learning  is also presented.Learning algorithm has a very important characteristic that as more is learned during FDA* algorithm’s problemsolving,with probability 1, Learning algorithm can make a heuristic evaluation function converge to an admissible one.