Wavelet networks are a class of neural network that take advantage of good localization properties of multi-resolution analysis and combine them with the approximation abilities of neural networks. This kind of networks uses wavelets as activation functions in the hidden layer and a type of back-propagation algorithm is used for its learning. However, the training procedure used for wavelet networks is based on the idea of continuous differentiable wavelets and some of the most powerful and used wavelets do not satisfy this property. In this paper we report an algorithm for initialising and training wavelet networks applied to the approximation of chaotic time series. The proposed algorithm which has its foundations on correlation analysis of signals allows the use of different types of wavelets, namely, Daubechies, Coiflets, and Symmlets. To show this, comparisons are made for chaotic time series approximation between the proposed approach and the typical wavelet network.