This does not come without a price: Aside from the complexity of searching two times in parallel, we have to decide which search tree to extend at each step; we have to be able to travel backwards from goal to initial state - which may not be possible without extra work; and we need an efficient way to find the intersection of the two search trees. This additional complexity means that the A* search algorithm is often a better choice if we have a reasonable heuristic.
Bi-directional search can use a heuristic as well. An admissible heuristic will also produce a shortest solution as was proven originally for the A* search algorithm.
A node to be expanded is selected from the frontier that has the least number of open nodes and which is most promising. Termination happens when such a node resides also in the other frontier. A descendant node's f-value must take into account the g-values of all open nodes at the other frontier. Hence node expansion is more costly than for A*. The collection of nodes to be visited can be smaller as outlined above. Thus one trades in less space for more computation. The 1977 reference showed that the bi-directional algorithm found solutions where A* had run out of space. Shorter paths were also found when non admissible heuristics were used. These tests were done on the 15-puzzle used by Ira Pohl.
Ira Pohl was the first one to design and implement a bi-directional heuristic search algorithm. A descendant node in a frontier was evaluated only regarding the target at the other side. He reported that the two search trees were missing each other and did not meet in the middle of the search space.