I've been using back-propagation to optimize neural networks without problems. But I've read in some books that genetic algorithms can be used to optimize an ANN. I want to know if its possible to use them as a learning algorithm.
Yes! There are essentially 2 ways of doing that: with fixed topology or evolving topology.
Fixed topology is the easiest. Just encode all weights as values in your chromossomes and evolve. You can use the network error as your fitness (to minimize, in this case). The disadvantage of this method is that you need to chose the network architecture manually. You can see this approach in action here.