Wavelet function based feed forward neural network architecture is proposed for forecasting of missed data. A set of wavelet functions offers a multi-resolution approximation in signal analysis and provides localization in spatial domain. Wavelet neural network (WNN) is developed using these properties of wavelet functions in the traditional feed forward network. This network has multi-layer architecture and its each neuron includes wavelet functions in the first layer. The second layer entries are wavelet coefficients corresponding to first layer outputs. Wavelet functions provide high convergence capability and faster learning process in the multilayer networks. WNN performance is tested for function approximation and missed data forecasting problems.