Based on the fuzzy neural network, an approach to early software reliability prediction is proposed. The reason of the software defects generated is analyzed and the factors which introduce the defects during software creation are presented. It is pointed out that fuzzy and using experience are the characteristics of the factors and the modeling in early software reliability prediction must consider them. The modeling method for early software reliability prediction is discussed and the detail steps for the approach are given. Future research works about this approach such as how to quantify the inputs of the model and how to train the fuzzy neural network are also mentioned.